Zeroth-order (ZO) optimization as the gradient-free method has become a powerful tool when the first-order gradient is unavailable or expensive to obtain, especially in decentralized learning scenarios where data and computational resources are distributed across multiple clients. There have been many efforts to analyze the optimization convergence rate of zeroth-order decentralized stochastic gradient descent (ZO-DSGD) algorithms. However, the generalization of these methods has not been well studied. In this paper, we provide a generalization analysis of ZO-DSGD with changing topology, where the clients run zeroth-order SGD with local data and communicate with each other according to time-varying topology. We systematically analyze the generalization error in convex, strongly convex, and non-convex cases. The obtained results in the convex and strongly convex cases with zeroth-order oracles recover the results of SGD. Moreover, the generalization bounds derived in non-convex cases align with that of DSGD. To capture the influence of communication topology on the generalization performance, we analyze local generalization bounds concerning local models held at different clients. The obtained results reflect the influence of the number of clients, local sample size, and topology on the generalization error. To the best of our knowledge, this is the first work that provides a generalization analysis of zeroth-order decentralized stochastic gradient descent methods and recovers the results of SGD.

Many positional encodings (PEs) are designed to exhibit long-term decay, based on an entrenched and long-standing inductive opinion: tokens farther away from the current position carry less relevant information. We argue that long-term decay is outdated in the era of LLMs, as LLMs are now applied to tasks demanding precise retrieval of in-context information from arbitrary positions. Firstly, we present empirical analyses on various PEs, demonstrating that models inherently learn attention with only a local-decay pattern while forming a U-shape pattern globally, contradicting the principle of long-term decay. Furthermore, we conduct a detailed analysis of rotary position encoding (RoPE, a prevalent relative positional encoding in LLMs), and found that the U-shape attention is caused by some learned components, which are also the key factor limiting RoPE’s expressiveness and extrapolation. Inspired by these insights, we propose High-frequency rotary Position Encoding (HoPE). HoPE replaces the specific components in RoPE with position-independent ones, retaining only high-frequency signals, which also breaks the principle of long-term decay in theory. HoPE achieves two major advantages: (1) Without constraints imposed by long-term decay, contradictory factors that limit attention optimization are removed. Thus, the model’s context awareness is enhanced. (2) HoPE exhibits greater robustness to the out-of-distribution behavior in attention patterns during extrapolation. The effectiveness of HoPE is validated through extensive experiments and with a large language model of up to 3 billion parameters.

An extensive high-quality instruction dataset is crucial for the instruction tuning process of Large Language Models (LLMs).Recent instruction expansion methods have demonstrated their capability to improve the quality and quantity of existing datasets, by prompting high-performance LLM to generate multiple new instructions from the original ones.However, existing methods focus on constructing multi-perspective prompts (e.g., increasing complexity or difficulty) to expand instructions, overlooking the "Fixed Thinking Pattern" issue of LLMs.This issue arises when repeatedly using the same set of prompts, causing LLMs to rely on a limited set of certain expressions to expand all instructions, potentially compromising the diversity of the final expanded dataset.This paper theoretically analyzes the causes of the "Fixed Thinking Pattern", and corroborates this phenomenon through multi-faceted empirical research.Furthermore, we propose a novel method based on dynamic prompt updating: Global Eye.Specifically, after a fixed number of instruction expansions, we analyze the statistical characteristics of newly generated instructions and then update the prompts.Experimental results show that our method enables LLaMA3-8B and LLaMA2-13B to surpass the performance of open-source LLMs and GPT3.5 across various metrics.

Low-rank adaptation (LoRA) and its variants have recently gained much interest due to their ability to avoid excessive inference costs. However, LoRA still encounters the following challenges: (1) Limitation of low-rank assumption; and (2) Its initialization method may be suboptimal. To this end, we propose PMSS(Pre-trained Matrices Skeleton Selection), which enables high-rank updates with low costs while leveraging semantic and linguistic information inherent in pre-trained weight. It achieves this by selecting skeletons from the pre-trained weight matrix and only learning a small matrix instead. Experiments demonstrate that PMSS outperforms LoRA and other fine-tuning methods across tasks with much less trainable parameters. We demonstrate its effectiveness, especially in handling complex tasks such as DROP benchmark(+3.4%/+5.9% on LLaMA2-7B/13B) and math reasoning (+12.89%/+5.61%/+3.11% on LLaMA2-7B, Mistral-7B and Gemma-7B of GSM8K).The code and model will be released soon.

Function calling significantly extends the application boundary of large language models (LLMs), where high-quality and diverse training data is critical for unlocking this capability. However, collecting and annotating real function-calling data is challenging, while synthetic data from existing pipelines often lack coverage and accuracy. In this paper, we present ToolACE, an automatic agentic pipeline designed to generate accurate, complex, and diverse tool-learning data, specifically tailored to the capabilities of LLMs. ToolACE leverages a novel self-evolution synthesis process to curate a comprehensive API pool of 26,507 diverse APIs. Dialogs are further generated through the interplay among multiple agents, under the guidance of a complexity evaluator. To ensure data accuracy, we implement a dual-layer verification system combining rule-based and model-based checks. We demonstrate that models trained on our synthesized data---even with only 8B parameters---achieve state-of-the-art performance, comparable to the latest GPT-4 models. Our model and a subset of the data are publicly available at https://huggingface.co/Team-ACE.

Large language models (LLMs) have made significant strides in code translation tasks. However, ensuring both the correctness and readability of translated code remains a challenge, limiting their effective adoption in real-world software development. In this work, we propose F2STrans, a function-to-style guiding paradigm designed to progressively improve the performance of LLMs in code translation. Our approach comprises two key stages: (1) Functional learning, which optimizes translation correctness using high-quality source-target code pairs mined from online programming platforms, and (2) Style learning, which improves translation readability by incorporating both positive and negative style examples. Additionally, we introduce a novel code translation benchmark that includes up-to-date source code, extensive test cases, and manually annotated ground-truth translations, enabling comprehensive functional and stylistic evaluations. Experiments on both our new benchmark and existing datasets demonstrate that our approach significantly improves code translation performance. Notably, our approach enables Qwen-1.5B to outperform prompt-enhanced Qwen-32B and GPT-4 on average across 20 diverse code translation scenarios.

Large Language Models (LLMs), built on Transformer architectures, exhibit remarkable generalization across a wide range of tasks. However, fine-tuning these models for specific tasks remains resource-intensive due to their extensive parameterization. In this paper, we explore two remarkable phenomena related to the attention mechanism during the fine-tuning of LLMs (where Wq, Wk, and Wv denote the weights of the query, key, and value layers, respectively). The first phenomenon, termed “Unequal Importance of Attention Matrices”, highlights the impact of fine-tuning different weight matrices. It shows that optimizing the Wv matrix yields significantly better performance than optimizing the Wk matrix. Fine-tuning only the Wq and Wv matrices is computationally efficient while delivering results comparable to, or even better than fine-tuning all three matrices (Wq, Wk, and Wv). The second phenomenon, “Attention Matrices with Customized Learning Rate Lead to Better Convergence”, emphasizes the importance of assigning distinct learning rates to these matrices. Specifically, a higher learning rate for the Wv matrix compared to Wq and Wk accelerates convergence and improves performance. Building on these insights, we propose a new strategy that improves fine-tuning efficiency in terms of both storage and time. Experimental results on benchmark datasets validate the effectiveness of this approach, supporting our theoretical findings. Our analysis lays the theoretical groundwork for configuring and improving algorithms in LLMs fine-tuning.

Menglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, Bin Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Qinzhuo Wu, Wei Liu, Jian Luan, Bin Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Early action prediction aiming to recognize which classes the actions belong to before they are fully conveyed is a very challenging task, owing to the insufficient discrimination information caused by the domain gaps among different temporally observed domains. Most of the existing approaches focus on using fully observed temporal domains to "guide" the partially observed domains while ignoring the discrepancies between the harder low-observed temporal domains and the easier highly observed temporal domains. The recognition models tend to learn the easier samples from the highly observed temporal domains and may lead to significant performance drops on low-observed temporal domains. Therefore, in this article, we propose a novel temporally observed domain contrastive network, namely, TODO-Net, to explicitly mine the discrimination information from the hard actions samples from the low-observed temporal domains by mitigating the domain gaps among various temporally observed domains for 3-D early action prediction. More specifically, the proposed TODO-Net is able to mine the relationship between the low-observed sequences and all the highly observed sequences belonging to the same action category to boost the recognition performance of the hard samples with fewer observed frames. We also introduce a temporal domain conditioned supervised contrastive (TD-conditioned SupCon) learning scheme to empower our TODO-Net with the ability to minimize the gaps between the temporal domains within the same action categories, meanwhile pushing apart the temporal domains belonging to different action classes. We conduct extensive experiments on two public 3-D skeleton-based activity datasets, and the results show the efficacy of the proposed TODO-Net.

Shihan Deng, Weikai Xu, Hongda Sun, Wei Liu, Tao Tan, Jianfeng Liu, Ang Li, Jian Luan, Bin Wang, Rui Yan, Shuo Shang. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Hongda Sun, Weikai Xu, Wei Liu, Jian Luan, Bin Wang, Shuo Shang, Ji-Rong Wen, Rui Yan. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Tool learning aims to extend the capabilities of large language models (LLMs) with external tools. A major challenge in tool learning is how to support a large number of tools, including unseen tools. To address this challenge, previous studies have proposed retrieving suitable tools for the LLM based on the user query. However, previously proposed methods do not consider the differences between seen and unseen tools, nor do they take the hierarchy of the tool library into account, which may lead to suboptimal performance for tool retrieval. Therefore, to address the aforementioned issues, we propose ToolRerank, an adaptive and hierarchy-aware reranking method for tool retrieval to further refine the retrieval results. Specifically, our proposed ToolRerank includes Adaptive Truncation, which truncates the retrieval results related to seen and unseen tools at different positions, and Hierarchy-Aware Reranking, which makes retrieval results more concentrated for single-tool queries and more diverse for multi-tool queries. Experimental results show that ToolRerank can improve the quality of the retrieval results, leading to better execution results generated by the LLM.

While Large Language Models (LLMs) have achieved remarkable success in various fields, the efficiency of training and inference remains a major challenge. To address this issue, we propose SUBLLM, short for Subsampling-Upsampling-Bypass Large Language Model, an innovative architecture that extends the core decoder-only framework by incorporating subsampling, upsampling, and bypass modules. The subsampling modules are responsible for shortening the sequence, while the upsampling modules restore the sequence length, and the bypass modules enhance convergence. In comparison to LLaMA, the proposed SUBLLM exhibits significant enhancements in both training and inference speeds as well as memory usage, while maintaining competitive few-shot performance. During training, SUBLLM increases speeds by 26% and cuts memory by 10GB per GPU. In inference, it boosts speeds by up to 37% and reduces memory by 1GB per GPU. The training and inference speeds can be enhanced by 34% and 52% respectively when the context window is expanded to 8192. Our code is available at https://github.com/XiaoMi/subllm.

Recently, tool-augmented LLMs have gained increasing attention.Given an instruction, toolaugmented LLMs can interact with various external tools in multiple rounds and provide a final answer.However, previous LLMs were trained on overly detailed instructions, which included API names or parameters, while real users would not explicitly mention these API details.This leads to a gap between trained LLMs and real-world scenarios.In addition, most works ignore whether the interaction process follows the instruction.To address these issues, we constructed a training dataset called MGToolBench, which contains statement and category-level instructions to better reflect realworld scenarios.In addition, we propose Tool-Planner, a two-stage reinforcement learning framework that utilizes path planning and two feedback mechanisms to enhance the LLM's task completion and instruction-following capabilities.Experimental results show that Tool-Planner significantly improves the Match Rate, Pass Rate and Win Rate by 26.8%, 20.2%, and 5.6% compared to the SOTA model.Human evaluation verifies that the multi-granularity instructions can better align with users' usage habits.Our data and code are available at https://github.com/XiaoMi/toolplanner.

In nested Named entity recognition (NER), entities are nested with each other, and thus requiring more data annotations to address. This leads to the development of few-shot nested NER, where the prevalence of pretrained language models with in-context learning (ICL) offers promising solutions. In this work, we introduce an effective and innovative ICL framework for the setting of few-shot nested NER. We improve the ICL prompt by devising a novel example demonstration selection mechanism, EnDe retriever. In EnDe retriever, we employ contrastive learning to perform three types of representation learning, in terms of semantic similarity, boundary similarity, and label similarity, to generate high-quality demonstration examples. Extensive experiments over three nested NER and four flat NER datasets demonstrate the efficacy of our system.

Event extraction (EE) is a critical direction in the field of information extraction, laying an important foundation for the construction of structured knowledge bases. EE from text has received ample research and attention for years, yet there can be numerous real-world applications that require direct information acquisition from speech signals, online meeting minutes, interview summaries, press releases, etc. While EE from speech has remained under-explored, this paper fills the gap by pioneering a SpeechEE, defined as detecting the event predicates and arguments from a given audio speech. To benchmark the SpeechEE task, we first construct a large-scale high-quality dataset. Based on textual EE datasets under the sentence, document, and dialogue scenarios, we convert texts into speeches through both manual real-person narration and automatic synthesis, empowering the data with diverse scenarios, languages, domains, ambiences, and speaker styles. Further, to effectively address the key challenges in the task, we tailor an E2E SpeechEE system based on the encoder-decoder architecture, where a novel Shrinking Unit module and a retrieval-aided decoding mechanism are devised. Extensive experimental results on all SpeechEE subsets demonstrate the efficacy of the proposed model, offering a strong baseline for the task. At last, being the first work on this topic, we shed light on key directions for future research. Our codes and the benchmark datasets are open at https://SpeechEE.github.io/

The development of AI in mental health is a growing field with potential global impact. Machine agents need to perceive users' mental states and respond empathically. Since mental states are often latent and implicit, building such chatbots requires both knowledge learning and knowledge utilization. Our work contributes to this by developing a chatbot that aims to recognize and empathetically respond to users' mental states. We introduce a Conditional Variational Autoencoders (CVAE)-based model that utilizes relation-aware commonsense knowledge to generate responses. This model, while not a replacement for professional mental health support, demonstrates promise in offering informative and empathetic interactions in a controlled environment. On the dataset EmpatheticDialogues, we compare with several SOTA methods and empirically validate the effectiveness of our approach on response informativeness and empathy exhibition. Detailed analysis is also given to demonstrate the learning capability as well as model interpretability. Our code is accessible at http://github.com/ChangyuChen347/COMET-VAE.

Recommender systems have been widely adopted in various online personal e-commerce applications for improving user experience. A long-standing challenge in recommender systems is how to provide accurate recommendation to users in cold-start situations where only a few user-item interactions can be observed. Recently, meta learning methods provide a promising solution, and most of them follow a way of parameter initialization where predictions can be fast adapted via multiple gradient descent steps. While these meta-learning recommenders promote model performance, how to derive a fundamental paradigm that enables both flexible approximations of complex user interaction distributions and effective task adaptations of global knowledge still remains a critical yet under-explored problem. To this end, we present the Flow-based Adaptive Neural Process (FANP), a new probabilistic meta-learning model where estimating the preference of each user is governed by an underlying stochastic process. Following an encoder-decoder generative framework, FANP is an effective few-shot function estimator that directly maps limited user interactions to a predictive distribution without complicated gradient updates. Through introducing a conditional normalization flow-based encoder, FANP can get rid of the model bias on latent variables and thereby derive more flexible variational distributions. Meanwhile, we propose a task-adaptive mechanism capturing the relevance of different tasks for improving adaptation ability of global knowledge. The learned task-specific and task-relevant representations are simultaneously exploited to generate the decoder parameters via a novel modulation-augmented hypernetwork. FANP is evaluated on both scenario-specific and user-specific cold-start recommendations on various real-world datasets. Extensive experimental results and detailed model analyses demonstrate that our model yields superior performance compared with multiple state-of-the-art meta-learning recommenders.

Video Anomaly detection, aiming to detect the abnormal behaviors in surveillance videos, is a challenging task since the anomalous events are diversified and complicated in different situations. And this makes it difficult to use one single static network architecture to extract useful information from diverse abnormal patterns. Therefore, in this article, we propose a novel Dynamic Self-Supervised Network (DSS-Net) to explore both spatial and temporal anomalous information. In our DSS-Net, we design a dynamic network to adaptively select suitable network architecture to extract latent features from different anomalous patterns and normal patterns. Specifically, we generate spatial and temporalpseudo-abnormaldata as the input of the dynamic network to conduct self-supervised learning. And we have a specific design on Hybrid Anomaly Dynamic Convolution (HAD-Conv) to extract features for diversified anomalous events adaptively. We utilize both normal and pseudo-abnormal data to encourage the dynamic network to mine the discriminative information. Furthermore, we design a feature separation loss to maximize the difference between the anomalous and normal videos. We evaluate our proposed method on four public anomaly detection datasets and achieve competitive results compared with the state-of-the-art approaches.

Graph classification plays an important role in a wide range of applications from biological prediction to social analysis. Traditional graph classification models built on graph kernels are hampered by the challenge of poor generalization as they are heavily dependent on the dedicated design of handcrafted features. Recently, graph neural networks (GNNs) become a new class of tools for analyzing graph data and have achieved promising performance. However, it is necessary to collect a large number of labeled graph data for training an accurate GNN, which is often unaffordable in real-world applications. Therefore, it is an open question to build GNNs under the condition of few-shot learning where only a few labeled graphs are available. In this article, we introduce a new Structure-aware Prototypical Neural Process (SPNP for short) for a few-shot graph classification. Specifically, at the encoding stage, SPNP first employs GNNs to capture graph structure information. Then, SPNP incorporates such structural priors into the latent path and the deterministic path for representing stochastic processes. At the decoding stage, SPNP uses a new prototypical decoder to define a metric space where unseen graphs can be predicted effectively. The proposed decoder, which contains a self-attention mechanism to learn the intraclass dependence between graphs, can enhance the class-level representations, especially for new classes. Furthermore, benefited from such a flexible encoding-decoding architecture, SPNP can directly map the context samples to a predictive distribution without any complicated operations used in previous methods. Extensive experiments demonstrate that SPNP achieves consistent and significant improvements over state-of-the-art methods. Further discussions are provided toward model efficiency and more detailed analysis.

Previous works on emotion recognition in conversation (ERC) follow a two-step paradigm, which can be summarized as first producing context-independent features via fine-tuning pretrained language models (PLMs) and then analyzing contextual information and dialogue structure information among the extracted features. However, we discover that this paradigm has several limitations. Accordingly, we propose a novel paradigm, i.e., exploring contextual information and dialogue structure information in the fine-tuning step, and adapting the PLM to the ERC task in terms of input text, classification structure, and training strategy. Furthermore, we develop our model BERT-ERC according to the proposed paradigm, which improves ERC performance in three aspects, namely suggestive text, fine-grained classification module, and two-stage training. Compared to existing methods, BERT-ERC achieves substantial improvement on four datasets, indicating its effectiveness and generalization capability. Besides, we also set up the limited resources scenario and the online prediction scenario to approximate real-world scenarios. Extensive experiments demonstrate that the proposed paradigm significantly outperforms the previous one and can be adapted to various scenes.

Hao Sun, Zhexin Zhang, Fei Mi, Yasheng Wang, Wei Liu, Jianwei Cui, Bin Wang, Qun Liu, Minlie Huang. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Zhibin Lan, Jiawei Yu, Xiang Li, Wen Zhang, Jian Luan, Bin Wang, Degen Huang, Jinsong Su. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Conventional knowledge distillation (KD) approaches are commonly employed to compress neural machine translation (NMT) models.However, they only obtain one lightweight student each time.Consequently, we have to conduct KD multiple times when different students are required at the same time, which could be resource-intensive.Additionally, these students are individually optimized, and thus lack interactions with each other, leading to their potential not being fully exerted.In this work, we propose a novel All-In-One Knowledge Distillation (AIO-KD) framework for NMT, which generates multiple satisfactory students at once.Under AIO-KD, we first randomly extract fewer-layer subnetworks from the teacher as the sample students.Then, we jointly optimize the teacher and these students, where the students simultaneously learn the knowledge from the teacher and interact with other students via mutual learning.When utilized, we re-extract the candidate students, satisfying the specifications of various devices.Particularly, we adopt carefully-designed strategies for AIO-KD: 1) we dynamically detach gradients to prevent poorly-performed students from negatively affecting the teacher during the knowledge transfer, which could subsequently impact other students; 2) we design a twostage mutual learning strategy, which alleviates the negative impacts of poorly-performed students on the early-stage student interactions.Extensive experiments and in-depth analyses on three benchmarks demonstrate the effectiveness and eco-friendliness of AIO-KD.Our source code is available at https://github. com/DeepLearnXMU/AIO-KD.

Recent studies on end-to-end speech translation(ST) have facilitated the exploration of multilingual end-to-end ST and end-to-end simultaneous ST. In this paper, we investigate end-to-end simultaneous speech translation in a one-to-many multilingual setting which is closer to applications in real scenarios. We explore a separate decoder architecture and a unified architecture for joint synchronous training in this scenario. To further explore knowledge transfer across languages, we propose an asynchronous training strategy on the proposed unified decoder architecture. A multi-way aligned multilingual end-to-end ST dataset was curated as a benchmark testbed to evaluate our methods. Experimental results demonstrate the effectiveness of our models on the collected dataset. Our codes and data are available at: https://github.com/XiaoMi/TED-MMST.

Simultaneous machine translation (SimulMT) models start translation before the end of the source sentence, making the translation monotonically aligned with the source sentence. However, the general full-sentence translation test set is acquired by offline translation of the entire source sentence, which is not designed for SimulMT evaluation, making us rethink whether this will underestimate the performance of SimulMT models. In this paper, we manually annotate a monotonic test set based on the MuST-C English-Chinese test set, denoted as SiMuST-C. Our human evaluation confirms the acceptability of our annotated test set. Evaluations on three different SimulMT models verify that the underestimation problem can be alleviated on our test set. Further experiments show that finetuning on an automatically extracted monotonic training set improves SimulMT models by up to 3 BLEU points.

Complex Knowledge Base Question Answering aims to answer a complex question over a Knowledge Base. A mainstream solution is based on information retrieval, which usually extracts a pivotal subgraph from entire Knowledge Base to locate candidate answers, and then determines the plausible answers with semantic matching between candidate answers and the question. However, such a paradigm can have two critical problems: 1) Complex Knowledge Base Question Answering can be sensitive to the subgraph, since a small subgraph may exclude the answers, while a large one may introduce a lot of noise; 2) directly deriving answers with semantic matching neglects the global topology in the Knowledge Base, which may limit the capability in answer reasoning. To tackle above challenges, we propose the Relation-Aware Subgraph Retrieval and Reasoning Network, where relations are emphasized to construct subgraphs and answer reasoning. Specifically, we present a Relation-Aware Subgraph Retrieval (RASR) method to initialize and prune subgraphs with the guidance of relation semantics. To compre-hensively understand the complex correlations between the question and candidate answers, we put forward a Relation-Aware Reasoning Network (RARN), which contains a text reasoning module focusing on the semantics understanding of the question and a graph reasoning module focusing on mining the topology between the topic entities and the answers. Experiments on two classical benchmark datasets show that our reasoning model outperforms the state-of-the-art results of Information Retrieval models. What's more, data statistical analysis on the subgraphs demonstrates the effectiveness of our proposed subgraph retrieval method.

In industry, web platforms such as Alibaba and Amazon often provide diverse services for users. Unsurprisingly, some developed services are data-rich, while some newly started services are data-scarce accompanied by severe data sparsity and cold-start problems. To alleviate the above problems and incubate new services easily, cross-domain recommendation (CDR) has attracted much attention from industrial and academic researchers. Generally, CDR aims to transfer rich user-item interaction information from related source domains (e.g., developed services) to boost recommendation quality of target domains (e.g., newly started services). For different scenarios, previous CDR methods can be roughly divided into two branches: (1) Data sparsity CDR fulfills user preference aided by other domain data to make intra-domain recommendations for users with few interactions, (2) Cold-start CDR projects user preference from other domain to make inter-domain recommendations for users with none interactions. In the past years, many outstanding CDR methods are emerged, however, to the best of our knowledge, none of them attempts to solve the two branches simultaneously. In this paper, we provide a unified framework, namely UniCDR, which can universally model different CDR scenarios by transferring the domain-shared information. Extensive experiments under the above 2 branches on 4 CDR scenarios and 6 public and large-scale industrial datasets demonstrate the effectiveness and universal ability of our UniCDR.

Abstract Semantic parsing maps natural language questions into logical forms, which can be executed against a knowledge base for answers. In real-world applications, the performance of a parser is often limited by the lack of training data. To facilitate zero-shot learning, data synthesis has been widely studied to automatically generate paired questions and logical forms. However, data synthesis methods can hardly cover the diverse structures in natural languages, leading to a large gap in sentence structure between synthetic and natural questions. In this paper, we propose a decomposition-based method to unify the sentence structures of questions, which benefits the generalization to natural questions. Experiments demonstrate that our method significantly improves the semantic parser trained on synthetic data (+7.9% on KQA and +8.9% on ComplexWebQuestions in terms of exact match accuracy). Extensive analysis demonstrates that our method can better generalize to natural questions with novel text expressions compared with baselines. Besides semantic parsing, our idea potentially benefits other semantic understanding tasks by mitigating the distracting structure features. To illustrate this, we extend our method to the task of sentence embedding learning, and observe substantial improvements on sentence retrieval (+13.1% for Hit@1).

Early activity prediction/recognition aims to recognize action categories before they are fully conveyed. Compared to full-length action sequences, partial video sequences only provide insufficient discrimination information, which makes predicting the class labels for some similar activities challenging, especially when only very few frames can be observed. To address this challenge, in this paper, we propose a novel meta negative network, namely, Magi-Net, that utilizes a contrastive learning scheme to alleviate the insufficiency of discriminative information. In our Magi-Net model, the positive samples are generated by augmenting an input anchor conditioned on all observation ratios, while the negative samples are selected from a trainable negative look-up memory (LUM) table, which stores the training samples and the corresponding misleading categories. Furthermore, a meta negative sample optimization strategy (MetaSOS) is proposed to boost the training of Magi-Net by encouraging the model to learn from the most informative negative samples via a meta learning scheme. Extensive experiments are conducted on several public skeleton-based activity datasets, and the results show the efficacy of the proposed Magi-Net model.

Early activity prediction, which aims to recognize class labels before actions are fully performed, is a very challenging task since partially observed action sequences contain insufficient class-discrimination information, and thus, many partial action sequences belonging to different categories may look very similar. Therefore, in this paper, we propose a novelguidance aware network (GA-Net)to boost the ability to distinguish different activities in diversified partially observed action sequences via metric learning. To mitigate the similarity problem of action segments at very early stages, the proposedguided metric learning module (GMLM)is able to encourage the feature extractor to mine class-discriminative information given partially observed sequences. Specifically, the GMLM is able to minimize the intraclass distance with a full-length guided direction approach and maximize the difference between interclass categories with different observation ratios. To enhance the similarities between the partial- and full-length sequences in the same action categories, we further introduce adistribution alignment module (DAM)that employs full-length guidance to pull the partially observed features closer to the global features. We evaluate our proposed method on three public human activity datasets and achieve competitive results compared with the state-of-the-art approaches.

Action recognition in video understanding is a challenging task, largely because of the complexity and difficulty in temporal modeling, making it suffer from motion information loss and misalignment of temporal attention in spatial dimensions. To overcome these difficulties, we propose a novel temporal modeling method calledAdjoint Enhancement Network(AE-Net), which can fully explore clues of motion and time in the long-range structure. The AE-Net mainly consists of two new modules: theInitial Adjoint Enhancement Module(IAE-Module), which deals with shallow features; and theGlobal Adjoint Enhancement Module(GAE-Module), which deals with global features. With a novel mechanism of parallel spatio-temporal convolution and difference fusion, the IAE-Module is to enhance the degree of motion transformation in shallow network features, exciting the potential of motion flow and avoiding motion information loss. The GAE-Module is proposed to improve the local temporal representation in long-range structures by feeding the enhanced feature differences into a spatial cascade module with residuals to resolve the misalignment of temporal attention in the spatial dimension.The experimental results show that our AE-Net can achieve state-of-the-art results in Something-Something V1, UCF-101 and HMDB-51 datasets.

Applying existing methods to emotional support conversation-which provides valuable assistance to people who are in need-has two major limitations: (a) they generally employ a conversation-level emotion label, which is too coarse-grained to capture user's instant mental state; (b) most of them focus on expressing empathy in the response(s) rather than gradually reducing user's distress.To address the problems, we propose a novel model MISC, which firstly infers the user's fine-grained emotional status, and then responds skillfully using a mixture of strategy.Experimental results on the benchmark dataset demonstrate the effectiveness of our method and reveal the benefits of fine-grained emotion understanding as well as mixed-up strategy modeling.Our code and data could be found in https: //github.com/morecry/MISC.

Cross-Domain Sequential Recommendation (CDSR) aims to predict future interactions based on user's historical sequential interactions from multiple domains. Generally, a key challenge of CDSR is how to mine precise cross-domain user preference based on the intra-sequence and inter-sequence item interactions. Existing works first learn single-domain user preference only with intra-sequence item interactions, and then build a transferring module to obtain cross-domain user preference. However, such a pipeline and implicit solution can be severely limited by the bottleneck of the designed transferring module, and ignores to consider inter-sequence item relationships.

Most existing methods on robust neural machine translation (NMT) construct adversarial examples by injecting noise into authentic examples and indiscriminately exploit two types of examples. They require the model to translate both the authentic source sentence and its adversarial counterpart into the identical target sentence within the same training stage, which may be a suboptimal choice to achieve robust NMT. In this paper, we first conduct a preliminary study to confirm this claim and further propose an Iterative Scheduled Data-switch Training Framework to mitigate this problem. Specifically, we introduce two training stages, iteratively switching between authentic and adversarial examples. Compared with previous studies, our model focuses more on just one type of examples at each single stage, which can better exploit authentic and adversarial examples, and thus obtaining a better robust NMT model. Moreover, we introduce an improved curriculum learning method with a sampling strategy to better schedule the process of noise injection. Experimental results show that our model significantly surpasses several competitive baselines on four translation benchmarks. Our source code is available at https://github.com/DeepLearnXMU/RobustNMT-ISDST.

Unsupervised cross-domain NER task aims to solve the issues when data in a new domain are fully-unlabeled. It leverages labeled data from source domain to predict entities in unlabeled target domain. Since training models on large domain corpus is time-consuming, in this paper, we consider an alternative way by introducing syntactic dependency structure. Such information is more accessible and can be shared between sentences from different domains. We propose a novel framework with dependency-aware GNN (DGNN) to learn these common structures from source domain and adapt them to target domain, alleviating the data scarcity issue and bridging the domain gap. Experimental results show that our method outperforms state-of-the-art methods.

Semantic parsing is a mainstream method of knowledge base question answering task that first generates a set of logical forms according to question and knowledge base (KB), and then selects the most matching one to get answers. However, existing selection methods are usually based on word-level matching, which cannot capture the structural information or solve the long-term dependency problem of entities. To solve this problem, we propose a syntax-based graph matching method, which explicitly models both question and logical form as graphs, and performs matching at both word-level and structure-level. The multi-level matching strategy not only captures the structural and semantic relations between entities but also explores their intrinsic relations. Such a design can greatly minimize the gap between question and most relevant knowledge from KB, and hence can reason more accurately. We conduct extensive experiments on several benchmarks and demonstrate the effectiveness of our proposed method.

Recommender systems have been widely deployed in many real-world applications, but usually suffer from the long-standing user cold-start problem. As a promising way, Cross-Domain Recommendation (CDR) has attracted a surge of inter-est, which aims to transfer the user preferences observed in the source domain to make recommendations in the target domain. Previous CDR approaches mostly achieve the goal by following the Embedding and Mapping (EMCDR) idea which attempts to learn a mapping function to transfer the pre-trained user repre-sentations (embeddings) from the source domain into the target domain. However, they pre-train the user/item representations independently for each domain, ignoring to consider both domain interactions simultaneously. Therefore, the biased pre-trained representations inevitably involve the domain-specific information which may lead to negative impact to transfer information across domains. In this work, we consider a key point of the CDR task: what information needs to be shared across domains? To achieve the above idea, this paper utilizes the information bottleneck (IB) principle, and proposes a novel approach termed as CDRIB to enforce the representations encoding the domain-shared information. To derive the unbiased representations, we devise two IB regularizers to model the cross-domain/in-domain user-item interactions simultaneously and thereby CDRIB could consider both domain interactions jointly for de-biasing. With an additional contrastive information regularizer, CDRIB can also capture cross-domain user-user correlations. In this way, those regularizers encourage the representations to encode the domain-shared information, which has the capability to make recommendations in both domains directly. To the best of our knowledge, this paper is the first work to capture the domain-shared information for cold-start users via variational information bottleneck. Empirical experiments illustrate that CDRIB outperforms the state-of-the-art approaches on four real-world cross-domain datasets, demonstrating the effectiveness of adopting the information bottleneck for CDR.

Real-world web applications such as Amazon and Netflix often provide services in multiple countries and regions (i.e., markets) around the world. Generally, different markets share similar item sets while containing different amounts of interaction data. Some markets are data-scarce and others are data-rich and leveraging those data from similar and data-rich auxiliary markets could enhance the data-scarce markets. In this paper, we explore multi-market recommendation (MMR), and propose a novel model called M$^3$Rec to improve all markets recommendation simultaneously. Since items play the role to bridge different markets, we argue that mining the similarities among items is the key point of MMR. Our M^3Rec preprocess two global item similarities: intra- and inter- market similarities. Specifically, we first learn the second-order intra-market similarity by adopting linear models with closed-form solutions, and then capture the high-order inter-market similarity by the random walk. Afterward, we incorporate the global item similarities for each local market. We conduct extensive experiments on five public available markets and compare with several state-of-the-art methods. Detailed experimental results demonstrate the effectiveness of our proposed method.

Data sparsity is a long-standing problem in recommender systems. To alleviate it, Cross-Domain Recommendation (CDR) has attracted a surge of interests, which utilizes the rich user-item interaction information from the related source domain to improve the performance on the sparse target domain. Recent CDR approaches pay attention to aggregating the source domain information to generate better user representations for the target domain. However, they focus on designing more powerful interaction encoders to learn both domains simultaneously, but fail to model different user preferences of different domains. Particularly, domain-specific preferences of the source domain usually provide useless information to enhance the performance in the target domain, and directly aggregating the domain-shared and domain-specific information together maybe hurts target domain performance. This work considers a key challenge of CDR: How do we transfer shared information across domains? Grounded in the information theory, we propose DisenCDR, a novel model to disentangle the domain-shared and domain-specific information. To reach our goal, we propose two mutual-information-based disentanglement regularizers. Specifically, an exclusive regularizer aims to enforce the user domain-shared representations and domain-specific representations encoding exclusive information. An information regularizer is to encourage the user domain-shared representations encoding predictive information for both domains. Based on them, we further derive a tractable bound of our disentanglement objective to learn desirable disentangled representations. Extensive experiments show that DisenCDR achieves significant improvements over state-of-the-art baselines on four real-world datasets.

In few-shot relational triple extraction (FS-RTE), one seeks to extract relational triples from plain texts by utilizing only few annotated samples. Recent work first extracts all entities and then classifies their relations. Such an entity-then-relation paradigm ignores the entity discrepancy between relations. To address it, we propose a novel task decomposition strategy, Relation-then-Entity, for FS-RTE. It first detects relations occurred in a sentence and then extracts the corresponding head/tail entities of the detected relations. To instantiate this strategy, we further propose a model, RelATE, which builds a dual-level attention to aggregate relation-relevant information to detect the relation occurrence and utilizes the annotated samples of the detected relations to extract the corresponding head/tail entities. Experimental results show that our model outperforms previous work by an absolute gain (18.98%, 28.85% in F1 in two few-shot settings).

Session-based recommendation (SBR) aims to predict a user's next clicked item based on an anonymous yet short interaction sequence. Previous SBR models, which rely only on the limited short-term transition information without utilizing extra valuable knowledge, have suffered a lot from the problem of data sparsity. This paper proposes a novel mirror graph enhanced neural model for session-based recommendation (MGS), to exploit item attribute information over item embeddings for more accurate preference estimation.

Conversational recommendation system aims to recommend appropriate items to user by directly asking preference on attributes or recommending item list. However, most of existing methods only employ the flat item and attribute relationship, and ignore the hierarchical relationship connected by the similar user which can provide more comprehensive information. And these methods usually use the user accepted attributes to represent the conversational history and ignore the hierarchical information of sequential transition in the historical turns. In this paper, we propose Hierarchical Information-aware Conversational Recommender (HICR) to model the two types of hierarchical information to boost the performance of CRS. Experiments conducted on four benchmark datasets verify the effectiveness of our proposed model.

Existing studies for multi-source neural machine translation (NMT) either separately model different source sentences or resort to the conventional single-source NMT by simply concatenating all source sentences. However, there exist two drawbacks in these approaches. First, they ignore the explicit word-level semantic interactions between source sentences, which have been shown effective in the embeddings of multilingual texts. Second, multiple source sentences are simultaneously encoded by an NMT model, which is unable to fully exploit the semantic information of each source sentence. In this paper, we explore multi-stage information interactions for multi-source NMT. Specifically, we first propose a multi-source NMT model that performs information interactions at the encoding stage. Its encoder contains multiple semantic interaction layers, each of which sequentially consists of (1) monolingual semantic interaction sub-layer, which is based on the self-attention mechanism and used to learn word-level monolingual contextual representations of source sentences, and (2) cross-lingual semantic interaction sub-layer, which leverages word alignments to perform fine-grained semantic transitions among hidden states of different source sentences. Furthermore, at the training stage, we introduce a mutual distillation based training framework, where single-source models and ours perform information interactions. Such framework can fully exploit the semantic information of each source sentence to enhance our model. Extensive experimental results on the WMT14 English-German-French dataset show our method exhibits significant improvements upon competitive baselines.

The key point in multimodal learning is to learn semantic alignment that finds the correspondence between sub-elements of instances from different modality data. Attention mechanism has shown its power in semantic alignment learning as it enables to densely associate sub-elements across different modalities. However, for each sub-element, existing attention aligns it with all the sub-elements from another modality, while most of them have no correspondence with it, i.e. irrelevant sub-elements. The irrelevant sub-elements will distract the semantic alignment if they are also attended. In this paper, we propose a novel focal attention mechanism to learn more accurate semantic alignment. The focal attention sparsely attends to a subset of sub-elements, which are identified as relevant ones according to their posterior probabilities given each sub-element from another modality. Based on the observation that relevant sub-elements mostly describe the same semantic, the posterior probability can precisely distinguish relevant and irrelevant ones by taking interactions within the same modality into consideration, such that relevant sub-elements get higher and closer posterior probabilities, while irrelevant ones get lower probabilities. Such a design learns better semantic alignment by preventing the interference of irrelevant sub-elements, and it facilitates subsequent multimodal tasks that demand semantic alignment. To validate the effectiveness of the focal attention, we conduct extensive experiments on image-text matching and text-to-image generation, and we propose a bidirectional and stacked version of focal attention for them, respectively. Experimental results on benchmarks show that the focal attention can significantly and consistently outperform state-of-the-arts.

In multi-turn dialog, utterances do not always take the full form of sentences (Carbonell 1983), which naturally makes understanding the dialog context more difficult. However, it is essential to fully grasp the dialog context to generate a reasonable response. Hence, in this paper, we propose to improve the response generation performance by examining the model's ability to answer a reading comprehension question, where the question is focused on the omitted information in the dialog. Enlightened by the multi-task learning scheme, we propose a joint framework that unifies these two tasks, sharing the same encoder to extract the common and task-invariant features with different decoders to learn task-specific features. To better fusing information from the question and the dialog history in the encoding part, we propose to augment the Transformer architecture with a memory updater, which is designed to selectively store and update the history dialog information so as to support downstream tasks. For the experiment, we employ human annotators to write and examine a large-scale dialog reading comprehension dataset. Extensive experiments are conducted on this dataset, and the results show that the proposed model brings substantial improvements over several strong baselines on both tasks. In this way, we demonstrate that reasoning can indeed help better response generation and vice versa. We release our large-scale dataset for further research.

Code generation aims to automatically generate a piece of code given an input natural language utterance. Currently, among dominant models, it is treated as a sequence-to-tree task, where a decoder outputs a sequence of actions corresponding to the pre-order traversal of an Abstract Syntax Tree. However, such a decoder only exploits the pre-order traversal based preceding actions, which are insufficient to ensure correct action predictions. In this paper, we first throughly analyze the context modeling difference between neural code generation models with different traversals based decodings (preorder traversal vs breadth-first traversal), and then propose to introduce a mutual learning framework to jointly train these models. Under this framework, we continuously enhance both two models via mutual distillation, which involves synchronous executions of two one-to-one knowledge transfers at each training step. More specifically, we alternately choose one model as the student and the other as its teacher, and require the student to fit the training data and the action prediction distributions of its teacher. By doing so, both models can fully absorb the knowledge from each other and thus could be improved simultaneously. Experimental results and in-depth analysis on several benchmark datasets demonstrate the effectiveness of our approach. We release our code at https://github.com/DeepLearnXMU/CGML.

Open Information Extraction (OpenIE) aims to discover textual facts from a given sentence.In essence, the facts contained in plain text are unordered.However, the popular Ope-nIE systems usually output facts sequentially in the way of predicting the next fact conditioned on the previous decoded ones, which enforce an unnecessary order on the facts and involve the error accumulation between autoregressive steps.To break this bottleneck, we propose MacroIE, a novel non-autoregressive framework for OpenIE.MacroIE firstly constructs a fact graph based on the table filling scheme, in which each node denotes a fact element, and an edge links two nodes that belong to the same fact.Then OpenIE can be reformulated as a non-parametric process of finding maximal cliques from the graph.It directly outputs the final set of facts in one go, thus getting rid of the burden of predicting fact order, as well as the error propagation between facts.Experiments conducted on two benchmark datasets show that our proposed model significantly outperforms current state-of-theart methods, beats the previous systems by as much as 5.7 absolute gain in F1 score.

Recent deep learning based methods have achieved impressive performance on paraphrase identification (PI), a fundamental NLP task, judging whether two sentences are semantically equivalent or not. However, their success heavily relies on massive labeled samples, which are time-consuming and expensive to obtain. To alleviate this problem, this study explores the effect of word alignment information (WAI), extracted by existing monolingual alignment tools, on deep PI baseline models. Apart from directly encoding WAI into fixed-size embeddings, we propose a novel auxiliary task so that the baselines can be pre-trained using a large amount of unlabeled in-domain data. Moreover, our proposed auxiliary task can also jointly train with the baselines, aiming to eliminate the overheads of preprocessing WAI at the test period. Experimental results verify that our methods can significantly outperform the deep PI baseline model.

Cross-domain slot filling is a challenging task in spoken language understanding due to the differences in text genre across domains. In this paper, we attempt to solve this task by exploiting the syntactic structures of user utterances, because these syntactic structures are actually accessible and can be shared between utterances from different domains. To this end, we propose a novel Syntactic Structure Encoder (SSE) module and incorporate it into a detection-prediction framework. SSE introduces graph convolutional network (GCN) to learn the common structures from multiple source domains, which are helpful to better adaptation on the target domain. Experimental results conducted on SNIPS dataset show that our model significantly outperforms the state-of-the-art approach in cross-domain slot filling. Specifically, our model outperforms the best model by ∼4% and ∼5% F1-scores under the 20-example and 50-example settings, respectively.

In this paper, we propose a robust neural machine translation (NMT) framework to deal with homophone errors. The framework consists of a homophone noise detector and a syllable-aware NMT model. The detector identifies potential homophone errors in a textual sentence and converts them into syllables to form a mixed sequence that is then fed into the syllable-aware NMT. Extensive experiments on Chinese→English translation demonstrate that the proposed method not only significantly outperforms baselines on noisy test sets with homophone noise, but also achieves substantial improvements over them on clean texts.

Multivariate Hawkes processes (MHPs) are classic methods to learn temporal patterns in event sequences of different entities. Traditional MHPs with explicit parametric intensity functions are friendly to model interpretability. However, recent Deep MHPs which employ various variants of recurrent neural networks are hardly to understand, albeit more expressive towards event sequences. The lack of model interpretability of Deep MHPs leads to a limited comprehension of complicated dynamics between events. To this end, we present a new Disentangled Deep Multivariate Hawkes Process $(\mathrm{D}^{2}$ MHP) to enhance model expressiveness and meanwhile maintain model interpretability. $\mathrm{D}^{2}$ MHP achieves state disentanglement by disentangling the latent representation of an event sequence into static and dynamic latent variables, and matches these latent variables to interpretable factors in the intensity function. Moreover, considering that an entity typically has multiple identities, $\mathrm{D}^{2}$ MHP further splits these latent variables into factorized representations, each of which is associated with a corresponding identity. Experiments on real-world datasets show that $\mathrm{D}^{2}$ MHP yields significant and consistent improvements over state-of-the-art baselines. We also demonstrate model interpretability via the detailed analysis.

Zhen Ke, Liang Shi, Songtao Sun, Erli Meng, Bin Wang, Xipeng Qiu. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Existing emotion-aware conversational models usually focus on controlling the response contents to align with a specific emotion class, whereas empathy is the ability to understand and concern the feelings and experience of others. Hence, it is critical to learn the causes that evoke the users' emotion for empathetic responding, a.k.a. emotion causes. To gather emotion causes in online environments, we leverage counseling strategies and develop an empathetic chatbot to utilize the causal emotion information. On a real-world online dataset, we verify the effectiveness of the proposed approach by comparing our chatbot with several SOTA methods using automatic metrics, expert-based human judgements as well as user-based online evaluation.

Bipartite graph embedding has recently attracted much attention due to the fact that bipartite graphs are widely used in various application domains. Most previous methods, which adopt random walk-based or reconstruction-based objectives, are typically effective to learn local graph structures. However, the global properties of bipartite graph, including community structures of homogeneous nodes and long-range dependencies of heterogeneous nodes, are not well preserved. In this paper, we propose a bipartite graph embedding called BiGI to capture such global properties by introducing a novel local-global infomax objective. Specifically, BiGI first generates a global representation which is composed of two prototype representations. BiGI then encodes sampled edges as local representations via the proposed subgraph-level attention mechanism. Through maximizing the mutual information between local and global representations, BiGI enables nodes in bipartite graph to be globally relevant. Our model is evaluated on various benchmark datasets for the tasks of top-K recommendation and link prediction. Extensive experiments demonstrate that BiGI achieves consistent and significant improvements over state-of-the-art baselines. Detailed analyses verify the high effectiveness of modeling the global properties of bipartite graph.

User cold-start recommendation is a long-standing challenge for recommender systems due to the fact that only a few interactions of cold-start users can be exploited. Recent studies seek to address this challenge from the perspective of meta learning, and most of them follow a manner of parameter initialization, where the model parameters can be learned by a few steps of gradient updates. While these gradient-based meta-learning models achieve promising performances to some extent, a fundamental problem of them is how to adapt the global knowledge learned from previous tasks for the recommendations of cold-start users more effectively.

Open Information Extraction (OIE), the task aimed at discovering all textual facts organized in the form of (subject, predicate, object) found within a sentence, has gained much attention recently. However, in some knowledge-driven applications such as question answering, we often have a target entity and hope to obtain its structured factual knowledge for better understanding, instead of extracting all possible facts aimlessly from the corpus. In this paper, we define a new task, namely Semi-Open Information Extraction (SOIE), to address this need. The goal of SOIE is to discover domain-independent facts towards a particular entity from general and diverse web text. To facilitate research on this new task, we propose a large-scale human-annotated benchmark called SOIED, consisting of 61,984 facts for 8,013 subject entities annotated on 24,000 Chinese sentences collected from the web search engine.

Graph Neural Networks (GNNs) have achieved state-of-the-art performance in many graph data analysis tasks. However, they still suffer from two limitations for graph representation learning. First, they exploit non-smoothing node features which may result in suboptimal embedding and degenerated performance for graph classification. Second, they only exploit neighbor information but ignore global topological knowledge. Aiming to overcome these limitations simultaneously, in this paper, we propose a novel, flexible, and end-to-end framework, Graph Smoothing Splines Neural Networks (GSSNN), for graph classification. By exploiting the smoothing splines, which are widely used to learn smoothing fitting function in regression, we develop an effective feature smoothing and enhancement module Scaled Smoothing Splines (S3) to learn graph embedding. To integrate global topological information, we design a novel scoring module, which exploits closeness, degree, as well as self-attention values, to select important node features as knots for smoothing splines. These knots can be potentially used for interpreting classification results. In extensive experiments on biological and social datasets, we demonstrate that our model achieves state-of-the-arts and GSSNN is superior in learning more robust graph representations. Furthermore, we show that S3 module is easily plugged into existing GNNs to improve their performance.

Non-autoregressive models generate target words in a parallel way, which achieve a faster decoding speed but at the sacrifice of translation accuracy.To remedy a flawed translation by non-autoregressive models, a promising approach is to train a conditional masked translation model (CMTM), and refine the generated results within several iterations.Unfortunately, such approach hardly considers the sequential dependency among target words, which inevitably results in a translation degradation.Hence, instead of solely training a Transformer-based CMTM, we propose a Self-Review Mechanism to infuse sequential information into it.Concretely, we insert a left-to-right mask to the same decoder of CMTM, and then induce it to autoregressively review whether each generated word from CMTM is supposed to be replaced or kept.The experimental results (WMT14 En↔De and WMT16 En↔Ro) demonstrate that our model uses dramatically less training computations than the typical CMTM, as well as outperforms several state-of-the-art non-autoregressive models by over 1 BLEU.Through knowledge distillation, our model even surpasses a typical left-to-right Transformer model, while significantly speeding up decoding.

Incorporating lexicons into character-level Chinese NER by lattices is proven effective to exploit rich word boundary information.Previous work has extended RNNs to consume lattice inputs and achieved great success.However, due to the DAG structure and the inherently unidirectional sequential nature, this method precludes batched computation and sufficient semantic interaction.In this paper, we propose PLTE, an extension of transformer encoder that is tailored for Chinese NER, which models all the characters and matched lexical words in parallel with batch processing.PLTE augments self-attention with positional relation representations to incorporate lattice structure.It also introduces a porous mechanism to augment localness modeling and maintain the strength of capturing the rich long-term dependencies.Experimental results show that PLTE performs up to 11.4 times faster than state-of-the-art methods while realizing better performance.We also demonstrate that using BERT representations further substantially boosts the performance and brings out the best in PLTE.

Dependency trees have been shown to be effective in capturing long-range relations between target entities.Nevertheless, how to selectively emphasize target-relevant information and remove irrelevant content from the tree is still an open problem.Existing approaches employing predefined rules to eliminate noise may not always yield optimal results due to the complexity and variability of natural language.In this paper, we present a novel architecture named Dynamically Pruned Graph Convolutional Network (DP-GCN), which learns to prune the dependency tree with rethinking in an end-to-end scheme.In each layer of DP-GCN, we employ a selection module to concentrate on nodes expressing the target relation by a set of binary gates and then augment the pruned tree with a pruned semantic graph to ensure the connectivity.After that, we introduce a rethinking mechanism to guide and refine the pruning operation by feeding back the high-level learned features repeatedly.Extensive experimental results demonstrate that our model achieves impressive performance compared to strong competitors.

Image-text matching has received growing interest since it bridges vision and language. The key challenge lies in how to learn correspondence between image and text. Existing works learn coarse correspondence based on object co-occurrence statistics, while failing to learn fine-grained phrase correspondence. In this paper, we present a novel Graph Structured Matching Network (GSMN) to learn fine-grained correspondence. The GSMN explicitly models object, relation and attribute as a structured phrase, which not only allows to learn correspondence of object, relation and attribute separately, but also benefits to learn fine-grained correspondence of structured phrase. This is achieved by node-level matching and structure-level matching. The node-level matching associates each node with its relevant nodes from another modality, where the node can be object, relation or attribute. The associated nodes then jointly infer fine-grained correspondence by fusing neighborhood associations at structure-level matching. Comprehensive experiments show that GSMN outperforms state-of-the-art methods on benchmarks, with relative Recall@1 improvements of nearly 7% and 2% on Flickr30K and MSCOCO, respectively. Code will be released at: https://github.com/CrossmodalGroup/GSMN.

Joint extraction of entities and relations aims to detect entity pairs along with their relations using a single model. Prior work typically solves this task in the extract-then-classify or unified labeling manner. However, these methods either suffer from the redundant entity pairs, or ignore the important inner structure in the process of extracting entities and relations. To address these limitations, in this paper, we first decompose the joint extraction task into two interrelated subtasks, namely HE extraction and TER extraction. The former subtask is to distinguish all head-entities that may be involved with target relations, and the latter is to identify corresponding tail-entities and relations for each extracted head-entity. Next, these two subtasks are further deconstructed into several sequence labeling problems based on our proposed span-based tagging scheme, which are conveniently solved by a hierarchical boundary tagger and a multi-span decoding algorithm. Owing to the reasonable decomposition strategy, our model can fully capture the semantic interdependency between different steps, as well as reduce noise from irrelevant entity pairs. Experimental results show that our method outperforms previous work by 5.2%, 5.9% and 21.5% (F1 score), achieving a new state-of-the-art on three public datasets.

More recently, Named Entity Recognition has achieved great advances aided by pre-training approaches such as BERT.However, current pre-training techniques focus on building language modeling objectives to learn a general representation, ignoring the named entityrelated knowledge.To this end, we propose a NER-specific pre-training framework to inject coarse-to-fine automatically mined entity knowledge into pre-trained models.Specifically, we first warm-up the model via an entity span identification task by training it with Wikipedia anchors, which can be deemed as general-typed entities.Then we leverage the gazetteer-based distant supervision strategy to train the model extract coarse-grained typed entities.Finally, we devise a self-supervised auxiliary task to mine the fine-grained named entity knowledge via clustering.Empirical studies on three public NER datasets demonstrate that our framework achieves significant improvements against several pre-trained baselines, establishing the new state-of-the-art performance on three benchmarks.Besides, we show that our framework gains promising results without using human-labeled training data, demonstrating its effectiveness in labelfew and low-resource scenarios.1

Graph neural networks (GNNs) have been successfully used to analyze non-Euclidean network data. Recently, there emerge a number of works to investigate the robustness of GNNs by adding adversarial noises into the graph topology, where gradient-based attacks are widely studied due to their inherent efficiency and high effectiveness. However, the gradient-based attacks often lead to sub-optimal results due to the discrete structure of graph data. To this end, we design a novel exploratory adversarial attack (termed as EpoAtk) to boost the gradient-based perturbations on graphs. The exploratory strategy in EpoAtk includes three phases, generation, evaluation and recombination, with the goal of sidesteping the possible misinformation that the maximal gradient provides. In experiments, EpoAtk is evaluated on benchmark datasets for the task of semi-supervised node classification in different attack settings. Experimental results demonstrate that the proposed method significantly outperforms the state-of-the-art attacks with the same attack budgets.

The lack of sufficient training data in many domains, poses a major challenge to the construction of domain-specific machine reading comprehension (MRC) models with satisfying performance. In this paper, we propose a novel iterative multi-source mutual knowledge transfer framework for MRC. As an extension of the conventional knowledge transfer with one-to-one correspondence, our framework focuses on the many-to-many mutual transfer, which involves synchronous executions of multiple many-to-one transfers in an iterative manner.Specifically, to update a target-domain MRC model, we first consider other domain-specific MRC models as individual teachers, and employ knowledge distillation to train a multi-domain MRC model, which is differentially required to fit the training data and match the outputs of these individual models according to their domain-level similarities to the target domain. After being initialized by the multi-domain MRC model, the target-domain MRC model is fine-tuned to match both its training data and the output of its previous best model simultaneously via knowledge distillation. Compared with previous approaches, our framework can continuously enhance all domain-specific MRC models by enabling each model to iteratively and differentially absorb the domain-shared knowledge from others. Experimental results and in-depth analyses on several benchmark datasets demonstrate the effectiveness of our framework.

Most Visual Question Answering (VQA) models suffer from the language prior problem, which is caused by inherent data biases. Specifically, VQA models tend to answer questions (e.g., what color is the banana?) based on the high-frequency answers (e.g., yellow) ignoring image contents. Existing approaches tackle this problem by creating delicate models or introducing additional visual annotations to reduce question dependency and strengthen image dependency. However, they are still subject to the language prior problem since the data biases have not been fundamentally addressed. In this paper, we introduce a self-supervised learning framework to solve this problem. Concretely, we first automatically generate labeled data to balance the biased data, and then propose a self-supervised auxiliary task to utilize the balanced data to assist the VQA model to overcome language priors. Our method can compensate for the data biases by generating balanced data without introducing external annotations. Experimental results show that our method achieves state-of-the-art performance, improving the overall accuracy from 49.50% to 57.59% on the most commonly used benchmark VQA-CP v2. In other words, we can increase the performance of annotation-based methods by 16% without using external annotations. Our code is available on GitHub.

With the rapid development of mobile Internet technology and the widespread use of mobile devices, it becomes much easier for people to express their opinions on social media. The openness and convenience of social media platforms provide a free expression for people but also cause new social problems. The widespread of false rumors on social media can bring about the panic of the public and damage personal reputation, which makes rumor automatic detection technology become particularly necessary. The majority of existing methods for rumor detection focus on mining effective features from text contents, user profiles, and patterns of propagation. Nevertheless, these methods do not take full advantage of global semantic relations of the text contents, which characterize the semantic commonality of rumors as a key factor for detecting rumors. In this paper, we construct a tweet-word-user heterogeneous graph based on the text contents and the source tweet propagations of rumors. A meta-path based heterogeneous graph attention network framework is proposed to capture the global semantic relations of text contents, together with the global structure information of source tweet propagations for rumor detection. Experiments on real-world Twitter data demonstrate the superiority of the proposed approach, which also has a comparable ability to detect rumors at a very early stage.

In this paper, we study the effect of label noise on deep learning models for paraphrase identification. Curriculum learning, a learning paradigm that learns easy samples first and then gradually proceeds with hard ones, has shown excellent results in dealing with label noise for deep neural networks (DNNs). However, most previous studies focus on image classification, and design their curriculum only based on training losses of samples, ignoring domain-specific knowledge. In this paper, we propose a predefined curriculum learning based framework, incorporating both training losses of samples and domain-specific knowledge, to train robust deep models for paraphrase identification (PI) with label noise. Through extensive experiments on two popular PI benchmarks, we show that 1) the performance of the deep paraphrase identification model can drop sharply at the case of severe label noise; 2) our approach can significantly improve generalization performance of deep networks trained on corrupted data especially at extremely high levels of label noise; 3) our method can outperform several state-of-the-art label corruption robust methods.

As a fundamental task in NLP, recognizing implicit discourse relations remains a challenging problem for years. One of the most important reasons is the limited amounts of annotated data. On one hand, most existing methods use multi-task methods to enlarge data. External datasets are often fully introduced to the training process which may lead to a negative transfer of noisy data. On the other hand, some previous researches on selecting data mostly focus on designing rules with heuristic methods. It should be difficult to cover all aspects of good samples. Another drawback is that data selection can hardly generally fit well in any other recognition model since it was specialized with a specific discourse relation classifier.In this paper, we propose a novel selective data enhanced model (SDE) based on reinforcement learning. Our model is a general framework, composed of two parts: 1) Discourse relation classifier is designed to identify relations, including multi-level representation module and relation recognition module. 2) Pseudo labeled data selector is designed to pick out data that can enhance the discourse relation classifier. We conduct joint learning alternately to optimize both of the classifier and the selector. Our model is able to expand data selectively. And classifier part can be replaced with any other complicated networks. To further exploit interact signals between arguments, we also present a multi-level representation based on BERT. Experiments show that our model achieves better performance than state-of-the-art methods.

While numerous approaches have been developed to embed graphs into either Euclidean or hyperbolic spaces, they do not fully utilize the information available in graphs, or lack the flexibility to model intrinsic complex graph geometry. To utilize the strength of both Euclidean and hyperbolic geometries, we develop a novel Geometry Interaction Learning (GIL) method for graphs, a well-suited and efficient alternative for learning abundant geometric properties in graph. GIL captures a more informative internal structural features with low dimensions while maintaining conformal invariance of each space. Furthermore, our method endows each node the freedom to determine the importance of each geometry space via a flexible dual feature interaction learning and probability assembling mechanism. Promising experimental results are presented for five benchmark datasets on node classification and link prediction tasks.

Image-text matching has received a large amount of interest since it associates different modalities and improves the understanding of image and natural language. It aims to retrieval semantic related images based on the given text query, and vice versa. Existing approaches have achieved much progress by projecting the image and text into a common space where data with different semantics can be distinguished. However, they process all the data points uniformly, while neglecting that data in a neighborhood are harder to distinguish due to their visual similarity or syntactic structural similarity. To address this issue, we propose a neighbor-aware network to image-text matching where an intra-attention module and neighbor-aware ranking loss are proposed to jointly distinguish data with different semantics, more importantly, semantic unrelated data in a neighborhood can be distinguished. The intra-attention attends to discriminative parts by comparing data with different semantics and magnifying difference between them, especially subtle difference between data in a neighborhood. The neighbor-aware ranking loss function utilizes the magnified difference to explicitly and effectively discriminate data in a neighborhood. We conduct extensive experiments on several benchmarks and show that the proposed approach significantly outperforms the state-of-the-art.

Cross-lingual Entity Alignment (CEA) aims at identifying entities with their counterparts in different language knowledge graphs. Knowledge embedding alignment plays an important role in CEA due to its advantages of easy implementation and run-time robustness. However, existing embedding alignment methods haven't considered the problem of embedding distribution alignment which refers to the alignment of spatial shapes of embedding spaces. To this end, we present a new Adversarial Knowledge Embedding framework (AKE for short) that jointly learns the representation, mapping and adversarial modules in an end-to-end manner. By reducing the discrepancy of embedding distributions, AKE can approximately preserve an isomorphism between source and target embeddings. In addition, we introduce two new orthogonality constraints into mapping to obtain the self-consistency and numerical stability of transformation. Experiments on real-world datasets demonstrate that our method significantly outperforms state-of-the-art baselines.

Heterogeneous graphs with different types of nodes and edges are ubiquitous and have immense value in many applications. Existing works on modeling heterogeneous graphs usually follow the idea of splitting a heterogeneous graph into multiple homogeneous subgraphs. This is ineffective in exploiting hidden rich semantic associations between different types of edges for large-scale multi-relational graphs. In this paper, we propose Relation Structure-Aware Heterogeneous Graph Neural Network (RSHN), a unified model that integrates graph and its coarsened line graph to embed both nodes and edges in heterogeneous graphs without requiring any prior knowledge such as metapath. To tackle the heterogeneity of edge connections, RSHN first creates a Coarsened Line Graph Neural Network (CL-GNN) to excavate edge-centric relation structural features that respect the latent associations of different types of edges based on coarsened line graph. After that, a Heterogeneous Graph Neural Network (H-GNN) is used to leverage implicit messages from neighbor nodes and edges propagating among nodes in heterogeneous graphs. As a result, different types of nodes and edges can enhance their embedding through mutual integration and promotion. Experiments and comparisons, based on semi-supervised classification tasks on large scale heterogeneous networks with over a hundred types of edges, show that RSHN significantly outperforms state-of-the-arts.

Benefiting from the excellent ability of neural networks on learning semantic representations, existing studies for entity linking (EL) have resorted to neural networks to exploit both the local mention-to-entity compatibility and the global interdependence between different EL decisions for target entity disambiguation. However, most neural collective EL methods depend entirely upon neural networks to automatically model the semantic dependencies between different EL decisions, which lack of the guidance from external knowledge. In this paper, we propose a novel end-to-end neural network with recurrent random-walk layers for collective EL, which introduces external knowledge to model the semantic interdependence between different EL decisions. Specifically, we first establish a model based on local context features, and then stack random-walk layers to reinforce the evidence for related EL decisions into high-probability decisions, where the semantic interdependence between candidate entities is mainly induced from an external knowledge base. Finally, a semantic regularizer that preserves the collective EL decisions consistency is incorporated into the conventional objective function, so that the external knowledge base can be fully exploited in collective EL decisions. Experimental results and in-depth analysis on various datasets show that our model achieves better performance than other state-of-the-art models. Our code and data are released at https://github.com/DeepLearnXMU/RRWEL.

Relation extraction studies the issue of predicting semantic relations between pairs of entities in sentences. Attention mechanisms are often used in this task to alleviate the inner-sentence noise by performing soft selections of words independently. Based on the observation that information pertinent to relations is usually contained within segments (continuous words in a sentence), it is possible to make use of this phenomenon for better extraction. In this paper, we aim to incorporate such segment information into neural relation extractor. Our approach views the attention mechanism as linear-chain conditional random fields over a set of latent variables whose edges encode the desired structure, and regards attention weight as the marginal distribution of each word being selected as a part of the relational expression. Experimental results show that our method can attend to continuous relational expressions without explicit annotations, and achieve the state-of-the-art performance on the large-scale TACRED dataset.

With the development of social media and the popularity of mobile devices, it becomes increasingly easy to post rumors and spread rumors on social media. Widespread rumors may cause public panic and negative impact on individuals, which makes the automatic detection of rumors become necessary. Most existing methods for automatic rumor detection focus on modeling features related to contents, users and propagation patterns based on feature engineering, but few work consider the existence of graph structural information in the user behavior. In this paper, we propose a model that leverages graph convolutional networks to capture user behavior effectively for rumor detection. Our model is composed of three modules: 1) a user encoder that models users attributes and behaviors based on graph convolutional networks to obtain user representation; 2) a propagation tree encoder, which encodes the structure of the rumor propagation tree as a vector with bridging the content semantics and propagation clues; 3) an integrator that integrates the output of the above modules to identify rumors. Experimental results on two public Twitter datasets show that our model achieves much better performance than the state-of-the-art methods.

Although distant supervised relation extraction has been widely used, its performance is weakened by the wrong labeling problem. During eliminating noise, many previous work failed to grasp the trade-off between long dependencies and computational complexity while using neural networks, and ignored the potential noise in external knowledge. In this paper, we propose a neural model incorporating dilated convolutional neural networks with soft entity type constraints to jointly tackle the above two issues. Rather than using traditional convolutional neural networks or recurrent neural networks, we propose dilated convolutional networks as the sentence encoder so as to capture long dependencies in large scale context and improve the robustness against local noise while keeping the efficiency of computation. Furthermore, we utilize entity types, which is considerd as external knowledge with noise, to denoise in relation classification by taking the constraints between entity types and relations into account and learning more precise entity types and attention weights simultaneously. Experimental results on the New York Times dataset show that, our model incorporating dilated convolutional networks with soft entity type constraints promotes distant supervised relation extraction and achieves state-of-art results compared with baselines.

Existing generative adversarial networks (GANs) with manifold regularization for semi-supervised learning (SSL) have shown promising performance in image generation and semi-supervised learning (SSL), which penalize the smoothness of classifier over data manifold based on the smoothness assumption. However, the smoothness assumption is valid for data points in high density region while not hold for data points in low density region, thus they tend to misclassify boundary instances in low density region. In this paper, we propose a novel discriminative regularization method for semi-supervised learning with conditional generative adversarial nets (CGANs). In our method, the discriminative information from class conditional data distribution captured by CGANs is utilized to improve the discrimination of classifier. Different from regular manifold regularization, the discriminative regularization encourages the classifier invariance to local perturbations on the sub-manifold of each cluster, and distinct classification outputs for data points in different clusters. Moreover, our method can be easily implemented via the stochastic approximation without constructing the Laplacian graph or computing the Jacobian of classifier explicitly. Experimental results on benchmark datasets show that our method can achieve competitive performance against previous advanced methods.

Learning semantic correspondence between image and text is significant as it bridges the semantic gap between vision and language. The key challenge is to accurately find and correlate shared semantics in image and text. Most existing methods achieve this goal by representing the shared semantic as a weighted combination of all the fragments (image regions or text words), where fragments relevant to the shared semantic obtain more attention, otherwise less. However, despite relevant ones contribute more to the shared semantic, irrelevant ones will more or less disturb it, and thus will lead to semantic misalignment in the correlation phase. To address this issue, we present a novel Bidirectional Focal Attention Network (BFAN), which not only allows to attend to relevant fragments but also diverts all the attention into these relevant fragments to concentrate on them. The main difference with existing works is they mostly focus on learning attention weight while our BFAN focus on eliminating irrelevant fragments from the shared semantic. The focal attention is achieved by preassigning attention based on inter-modality relation, identifying relevant fragments based on intra-modality relation and reassigning attention. Furthermore, the focal attention is jointly applied in both image-to-text and text-to-image directions, which enables to avoid preference to long text or complex image. Experiments show our simple but effective framework significantly outperforms state-of-the-art, with relative [email protected] gains of 2.2% on both Flicr30K and MSCOCO benchmarks.

Embedding knowledge graphs (KGs) into continuous vector spaces is a focus of current research. Combining such an embedding model with logic rules has recently attracted increasing attention. Most previous attempts made a one-time injection of logic rules, ignoring the interactive nature between embedding learning and logical inference. And they focused only on hard rules, which always hold with no exception and usually require extensive manual effort to create or validate. In this paper, we propose Rule-Guided Embedding (RUGE), a novel paradigm of KG embedding with iterative guidance from soft rules. RUGE enables an embedding model to learn simultaneously from 1) labeled triples that have been directly observed in a given KG, 2) unlabeled triples whose labels are going to be predicted iteratively, and 3) soft rules with various confidence levels extracted automatically from the KG. In the learning process, RUGE iteratively queries rules to obtain soft labels for unlabeled triples, and integrates such newly labeled triples to update the embedding model. Through this iterative procedure, knowledge embodied in logic rules may be better transferred into the learned embeddings. We evaluate RUGE in link prediction on Freebase and YAGO. Experimental results show that: 1) with rule knowledge injected iteratively, RUGE achieves significant and consistent improvements over state-of-the-art baselines; and 2) despite their uncertainties, automatically extracted soft rules are highly beneficial to KG embedding, even those with moderate confidence levels. The code and data used for this paper can be obtained from https://github.com/iieir-km/RUGE.

Embedding knowledge graphs (KGs) into continuous vector spaces is a focus of current research.Early works performed this task via simple models developed over KG triples.Recent attempts focused on either designing more complicated triple scoring models, or incorporating extra information beyond triples.This paper, by contrast, investigates the potential of using very simple constraints to improve KG embedding.We examine non-negativity constraints on entity representations and approximate entailment constraints on relation representations.The former help to learn compact and interpretable representations for entities.The latter further encode regularities of logical entailment between relations into their distributed representations.These constraints impose prior beliefs upon the structure of the embedding space, without negative impacts on efficiency or scalability.Evaluation on WordNet, Freebase, and DBpedia shows that our approach is simple yet surprisingly effective, significantly and consistently outperforming competitive baselines.The constraints imposed indeed improve model interpretability, leading to a substantially increased structuring of the embedding space.Code and data are available at https://github.com/i ieir-km/ComplEx-NNE_AER.

Learning to hash has proven to be an effective solution for indexing high-dimensional data by projecting them to similarity-preserving binary codes. However, most existing methods end up the learning scheme with a binarization stage, i.e. binary quantization, which inevitably destroys the neighborhood structure of original data. As a result, those methods still suffer from great similarity loss and result in unsatisfactory indexing performance. In this paper we propose a novel hashing model, namely Post Tuned Hashing (PTH), which includes a new post-tuning stage to refine the binary codes after binarization. The post-tuning seeks to rebuild the destroyed neighborhood structure, and hence significantly improves the indexing performance. We cast the post-tuning into a binary quadratic optimization framework and, despite its NP-hardness, give a practical algorithm to efficiently obtain a high-quality solution. Experimental results on five noted image benchmarks show that our PTH improves previous state-of-the-art methods by 13-58% in mean average precision.

Knowledge graphs (KGs) are often significantly incomplete, necessitating a demand for KG completion. Path-based relation inference is one of the most important approaches to this task. Traditional methods treat each path between entity pairs as an atomic feature, thus inducing sparsity. Recently, neural network models solve this problem by decomposing a path as the sequence of relations in the path, before modelling path representations with Recurrent Neural Network (RNN) architectures. In cases there are multiple paths between an entity pair, state-of-the-art neural models either select only one path, or make usage of simple score pooling methods like Top-K, Average, LogSumExp. Unfortunately, none of these methods can model the scenario where relations can only be inferred by considering multiple informative paths collectively. In this paper, we propose a novel path-based relation inference model that learns entity pair representations with attentive path combination. Given an entity pair and a set of paths connecting the pair, our model allows for integrating information from each informative path, and form a dynamic entity pair representation for each query relation. We empirically evaluate the proposed method on a real-world dataset. Experimental results show that the proposed model achieves better performance than state-of-the-art path-based relation inference methods.

In this paper, we formally define the problem of representing and leveraging abstract event causality to power downstream applications. We propose a novel solution to this problem, which build an abstract causality network and embed the causality network into a continuous vector space. The abstract causality network is generalized from a specific one, with abstract event nodes represented by frequently co-occurring word pairs. To perform the embedding task, we design a dual cause-effect transition model. Therefore, the proposed method can obtain general, frequent, and simple causality patterns, meanwhile, simplify event matching. Given the causality network and the learned embeddings, our model can be applied to a wide range of applications such as event prediction, event clustering and stock market movement prediction. Experimental results demonstrate that 1) the abstract causality network is effective for discovering high-level causality rules behind specific causal events; 2) the embedding models perform better than state-of-the-art link prediction techniques in predicting events; and 3) the event causality embedding is an easy-to-use and sophisticated feature for downstream applications such as stock market movement prediction.

This paper considers the problem of embedding Knowledge Graphs (KGs) consisting of entities and relations into low-dimensional vector spaces. Most of the existing methods perform this task based solely on observed facts. The only requirement is that the learned embeddings should be compatible within each individual fact. In this paper, aiming at further discovering the intrinsic geometric structure of the embedding space, we proposeSemantically Smooth Embedding(SSE). The key idea of SSE is to take full advantage of additional semantic information and enforce the embedding space to be semantically smooth, i.e., entities belonging to the same semantic category will lie close to each other in the embedding space. Two manifold learning algorithms Laplacian Eigenmaps and Locally Linear Embedding are used to model the smoothness assumption. Both are formulated as geometrically based regularization terms to constrain the embedding task. Two lines of embedding strategies are tested, i.e., strategies based on latent distance models and strategies based on tensor factorization techniques. We empirically evaluate SSE on two benchmark tasks of link prediction and triple classification, and achieve significant and consistent improvements over state-of-the-art methods. The results demonstrate the superiority and generality of SSE.

Knowledge graph (KG) embedding is to embed components of a KG including entities and relations into continuous vector spaces, so as to simplify the manipulation while preserving the inherent structure of the KG. It can benefit a variety of downstream tasks such as KG completion and relation extraction, and hence has quickly gained massive attention. In this article, we provide a systematic review of existing techniques, including not only the state-of-the-arts but also those with latest trends. Particularly, we make the review based on the type of information used in the embedding task. Techniques that conduct embedding using only facts observed in the KG are first introduced. We describe the overall framework, specific model design, typical training procedures, as well as pros and cons of such techniques. After that, we discuss techniques that further incorporate additional information besides facts. We focus specifically on the use of entity types, relation paths, textual descriptions, and logical rules. Finally, we briefly introduce how KG embedding can be applied to and benefit a wide variety of downstream tasks such as KG completion, relation extraction, question answering, and so forth.

Knowledge bases (KBs) are often greatly incomplete, necessitating a demand for KB completion. The path ranking algorithm (PRA) is one of the most promising approaches to this task. Previous work on PRA usually follows a single-task learning paradigm, building a prediction model for each relation independently with its own training data. It ignores meaningful associations among certain relations, and might not get enough training data for less frequent relations. This paper proposes a novel multi-task learning framework for PRA, referred to as coupled PRA (CPRA). It first devises an agglomerative clustering strategy to automatically discover relations that are highly correlated to each other, and then employs a multi-task learning strategy to effectively couple the prediction of such relations. As such, CPRA takes into account relation association and enables implicit data sharing among them. We empirically evaluate CPRA on benchmark data created from Freebase. Experimental results show that CPRA can effectively identify coherent clusters in which relations are highly correlated. By further coupling such relations, CPRA significantly outperforms PRA, in terms of both predictive accuracy and model interpretability.

Distant supervision is an efficient approach that automatically generates labeled data for relation extraction (RE). Traditional distantly supervised RE systems rely heavily on handcrafted features, and hence suffer from error propagation. Recently, a neural network architecture has been proposed to automatically extract features for relation classification. However, this approach follows the traditional expressed-at-least-once assumption, and fails to make full use of information across different sentences. Moreover, it ignores the fact that there can be multiple relations holding between the same entity pair. In this paper, we propose a multi-instance multi-label convolutional neural network for distantly supervised RE. It first relaxes the expressed-at-least-once assumption, and employs cross-sentence max-pooling so as to enable information sharing across different sentences. Then it handles overlapping relations by multi-label learning with a neural network classifier. Experimental results show that our approach performs significantly and consistently better than state-of-the-art methods.

Embedding knowledge graphs into continuous vector spaces has recently attracted increasing interest.Most existing methods perform the embedding task using only fact triples.Logical rules, although containing rich background information, have not been well studied in this task.This paper proposes a novel method of jointly embedding knowledge graphs and logical rules.The key idea is to represent and model triples and rules in a unified framework.Specifically, triples are represented as atomic formulae and modeled by the translation assumption, while rules represented as complex formulae and modeled by t-norm fuzzy logics.Embedding then amounts to minimizing a global loss over both atomic and complex formulae.In this manner, we learn embeddings compatible not only with triples but also with rules, which will certainly be more predictive for knowledge acquisition and inference.We evaluate our method with link prediction and triple classification tasks.Experimental results show that joint embedding brings significant and consistent improvements over stateof-the-art methods.Particularly, it enhances the prediction of new facts which cannot even be directly inferred by pure logical inference, demonstrating the capability of our method to learn more predictive embeddings.

Shu Guo, Quan Wang, Bin Wang, Lihong Wang, Li Guo. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2015.

We consider the problem of embedding knowledge graphs (KGs) into continuous vector spaces.Existing methods can only deal with explicit relationships within each triple, i.e., local connectivity patterns, but cannot handle implicit relationships across different triples, i.e., contextual connectivity patterns.This paper proposes context-dependent KG embedding, a twostage scheme that takes into account both types of connectivity patterns and obtains more accurate embeddings.We evaluate our approach on the tasks of link prediction and triple classification, and achieve significant and consistent improvements over state-of-the-art methods.

Various binary features have been recently proposed in literature, aiming at improving the computational efficiency and storage efficiency of image retrieval applications. However, the most common way of using binary features is voting strategy based on brute-force matching, since binary features are discrete data points distributed in Hamming space, so that models based on clustering such as BoW are unsuitable for them. Although indexing mechanism substantially decreases the time cost, the brute-force matching strategy becomes a bottleneck that restricts the performance of binary features. To address this issue, we propose a simple but effective method, namely COIP (Coding by Order-independent Projection), which projects binary features into a binary code of limited bits. As a result, each image is represented by one single binary code that can be indexed for computational and storage efficiency. We prove that the similarity between the COIP codes of two images with probability proportional to the ratio of their matched features. A comprehensive evaluation with several state-of-the-art binary features is performed on benchmark dataset. Experimental results reveal that for binary feature based image retrieval, our approach improves the storage/time efficiency by one/two orders of magnitude, while the retrieval performance remains almost unchanged.

Click prediction is one of the fundamental problems in sponsored search. Most of existing studies took advantage of machine learning approaches to predict ad click for each event of ad view independently. However, as observed in the real-world sponsored search system, user's behaviors on ads yield high dependency on how the user behaved along with the past time, especially in terms of what queries she submitted, what ads she clicked or ignored, and how long she spent on the landing pages of clicked ads, etc. Inspired by these observations, we introduce a novel framework based on Recurrent Neural Networks (RNN). Compared to traditional methods, this framework directly models the dependency on user's sequential behaviors into the click prediction process through the recurrent structure in RNN. Large scale evaluations on the click-through logs from a commercial search engine demonstrate that our approach can significantly improve the click prediction accuracy, compared to sequence-independent approaches.

Estimating questions ’ difficulty levels is an important task in community question answering (CQA) services. Previous stud-ies propose to solve this problem based on the question-user comparisons extract-ed from the question answering threads. However, they suffer from data sparseness problem as each question only gets a lim-ited number of comparisons. Moreover, they cannot handle newly posted question-s which get no comparisons. In this pa-per, we propose a novel question difficul-ty estimation approach called Regularized Competition Model (RCM), which natu-rally combines question-user comparisons and questions ’ textual descriptions into a unified framework. By incorporating tex-tual information, RCM can effectively deal with data sparseness problem. We further employ a K-Nearest Neighbor approach to estimate difficulty levels of newly post-ed questions, again by leveraging textu-al similarities. Experiments on two pub-licly available data sets show that for both well-resolved and newly-posted question-s, RCM performs the estimation task sig-nificantly better than existing methods, demonstrating the advantage of incorpo-rating textual information. More interest-ingly, we observe that RCMmight provide an automatic way to quantitatively mea-sure the knowledge levels of words. 1

With the development of microblog services, tens of thousands of messages are produced every day and recommending useful messages according to users' interest is recognized as an effective way to overcome the information overload problem. Collaborative filtering which rooted from recommender system has been utilized for microblog recommendation, where social relationship information can help improve the recommendation performance. However, most of existing methods only consider the static relationship, i.e. the following relationship, which totally ignores the relationship conveyed by users' repost behaviors. To explore the effects of behavior based relationship on recommendation, we propose an Interaction Based Collaborative Filtering (IBCF) approach. Specifically, we first use topic model to analyze users' interactive behaviors and measure the topic-specific relationship strength, then we incorporate the relationship factor into the matrix factorization framework. Experimental results show that compared to the current popular social recommendation methods, IBCF can achieve better performance on the MAP and NDCG evaluation measures, and have better interpretability for the recommended results.

Caching query results is an efficient technique for Web search engines. Admission policy can prevent infrequent queries from taking space of more frequent queries in the cache. In this paper we present two novel admission policies tailored for query results cache. These policies are based on query results prefetching information. We also propose a demote operation for the query results cache to improve the cache hit ratio. We then use a trace of over 5 million queries to evaluate our admission policies, as well as traditional policies. Experimental results show that our prefetch-aware admission policies can achieve hit ratios better than state-of-the-art admission policies.

This paper focuses on the problem of Question Routing (QR) in Community Question Answering (CQA), which aims to route newly posted questions to the potential answerers who are most likely to answer them. Traditional methods to solve this problem only consider the text similarity features between the newly posted question and the user profile, while ignoring the important statistical features, including the question-specific statistical feature and the user-specific statistical features. Moreover, traditional methods are based on unsupervised learning, which is not easy to introduce the rich features into them. This paper proposes a general framework based on the learning to rank concepts for QR. Training sets consist of triples (q, asker, answerers) are first collected. Then, by introducing the intrinsic relationships between the asker and the answerers in each CQA session to capture the intrinsic labels/orders of the users about their expertise degree of the question q, two different methods, including the SVM-based and RankingSVM-based methods, are presented to learn the models with different example creation processes from the training set. Finally, the potential answerers are ranked using the trained models. Extensive experiments conducted on a real world CQA dataset from Stack Overflow show that our proposed two methods can both outperform the traditional query likelihood language model (QLLM) as well as the state-of-the-art Latent Dirichlet Allocation based model (LDA). Specifically, the RankingSVM-based method achieves statistical significant improvements over the SVM-based method and has gained the best performance.

The major challenge for Question Retrieval (QR) in Community Question Answering (CQA) is the lexical gap between the queried question and the historical questions. This paper proposes a novel Question-Answer Topic Model (QATM) to learn the latent topics aligned across the question-answer pairs to alleviate the lexical gap problem, with the assumption that a question and its paired answer share the same topic distribution. Experiments conducted on a real world CQA dataset from Yahoo! Answers show that combining both parts properly can get more knowledge than each part or both parts in a simple mixing way and combining our QATM with the state-of-the-art translation-based language model, where the topic and translation information is learned from the question-answer pairs at two different grained semantic levels respectively, can significantly improve the QR performance.

Query results caching is an efficient technique for Web search engines. In this paper we present User-Aware Cache, a novel approach tailored for query results caching, that is based on user characteristics. We then use a trace of around 30 million queries to evaluate User-Aware Cache, as well as traditional methods and theoretical upper bounds. Experimental results show that this approach can achieve hit ratios better than state-of-the-art methods.

Question recommendation that automatically recommends a new question to suitable users to answer is an appealing and challenging problem in the research area of Community Question Answering (CQA). Unlike in general recommender systems where a user has only a single role, each user in CQA can play two different roles (dual roles) simultaneously: as an asker and as an answerer. To the best of our knowledge, this paper is the first to systematically investigate the distinctions between the two roles and their different influences on the performance of question recommendation in CQA. Moreover, we propose a Dual Role Model (DRM) to model the dual roles of users effectively. With different indepen-dence assumptions, two variants of DRM are achieved. Finally, we present the DRM based approach to question recommendation which provides a mechanism for naturally integrating the user relation between the answerer and the asker with the content re-levance between the answerer and the question into a uni-fied probabilistic framework. Experiments using a real-world data crawled from Yahoo! Answers show that: (1) there are evident distinctions between the two roles of users in CQA. Additionally, the answerer role is more effective than the asker role for modeling candidate users in question recommendation; (2) compared with baselines utilizing a single role or blended roles based methods, our DRM based approach consistently and significantly improves the performance of question recommendation, demonstrating that our approach can model the user in CQA more reasonably and precisely.

Social annotations provide additional document description contributed by online users and they have been explored for improving search performance. However, most existing methods need offline analysis of the whole tagged corpus, which is computationally expensive and cannot fit specific queries well. In this paper, we propose to use tags for document re-ranking. Specifically, we first estimate document similarity by combining words and tags and then adjust the document ranks with the assumption that similar documents should have similar retrieval scores. On similarity estimation, we present a new feature extraction method, called CRMF, from which document similarity can be derived. The CRMF can integrate the content and relation properties of multiple views and mine their correspondence. Besides, it does not require that all the documents to have tags. We tested the proposed approach on collections which are derived from Clue Web and contain Delicious tags. The experimental results demonstrate the effectiveness of tags on document re-ranking, where CRMF is significantly better than other state-of-the-art methods using tags.

With the rapid development of the information technology, there exists the difficulty in deploying state-of-the-art retrieval models in environments such as peer-to-peer networks and pervasive computing, where it is expensive or even infeasible to maintain the global statistics. To this end, this paper presents an investigation in the validity of different statistical assumptions of term distributions. Based on the findings in this investigation, a variety of weighting models, called NG (standing for "no global statistics") models, are derived from the Divergence from Randomness framework, in which only the within-document statistics are used in the relevance weighting. Compared to the state-of-the-art weighting models in extensive experiments on various standard TREC test collections, our proposed NG models can provide acceptable retrieval performance in ad-hoc search, without the use of global statistics.

User generated social annotations provide extra information for describing document contents. In this paper, we propose an effective method to model the categorization property of social annotations and explore the potential of combining it with classical language models for improving retrieval performance. Specifically, a novel TR-LDA model is presented to take annotations as an additional source for generating document contents apart from the document itself. We provide strategies for representing and weighting the categorization property and develop an efficient inference algorithm, where space saving is taken into account. Experiments are carried out on synthetic datasets, where documents and queries come from the standard evaluation conference TREC and annotations come from the website Delicious.com. Our results demonstrate the effectiveness of the proposed method on the ad-hoc retrieval task, which significantly outperforms state-of-art baselines.

Most existing query expansion approaches for ad-hoc retrieval adopt overly simplistic textual representations that treat documents as bags of words and ignore inherent document structure. These simple representations often lead to incorrect independence assumptions in the proposed approaches and result in limited retrieval effectiveness. In this paper, we propose a novel query expansion technique that models the various types of dependencies that exist between original query terms and expansion terms within a robust, unified framework. The proposed model is called Hierarchical Markov random fields (HMRFs), based on Latent Concept Expansion (LCE). By exploiting implicit (or explicit) hierarchical structure within documents, HMRFs can incorporate hierarchical interactions which are important for modeling term dependencies in an efficient manner. Our rigorous experimental evaluation carried out using several TREC data sets shows that our proposed query expansion technique consistently and significantly outperforms the current state-of-the-art query expansion approaches, including relevance-based language models and LCE.

Pseudo-relevance feedback (PRF) via query-expansion has been proven to be e®ective in many information retrieval (IR) tasks. In most existing work, the top-ranked documents from an initial search are assumed to be relevant and used for PRF. One problem with this approach is that one or more of the top retrieved documents may be non-relevant, which can introduce noise into the feedback process. Besides, existing methods generally do not take into account the significantly different types of queries that are often entered into an IR system. Intuitively, Wikipedia can be seen as a large, manually edited document collection which could be exploited to improve document retrieval effectiveness within PRF. It is not obvious how we might best utilize information from Wikipedia in PRF, and to date, the potential of Wikipedia for this task has been largely unexplored. In our work, we present a systematic exploration of the utilization of Wikipedia in PRF for query dependent expansion. Specifically, we classify TREC topics into three categories based on Wikipedia: 1) entity queries, 2) ambiguous queries, and 3) broader queries. We propose and study the effectiveness of three methods for expansion term selection, each modeling the Wikipedia based pseudo-relevance information from a different perspective. We incorporate the expansion terms into the original query and use language modeling IR to evaluate these methods. Experiments on four TREC test collections, including the large web collection GOV2, show that retrieval performance of each type of query can be improved. In addition, we demonstrate that the proposed method out-performs the baseline relevance model in terms of precision and robustness.

Tag services have recently become one of the most popular Internet services on the World Wide Web. Due to the fact that a Web page can be associate with multiple tags, previous research on tag recommendation mainly focuses on improving its accuracy or efficiency through multi-label learning algorithms. However, as a Web page can also be split into multiple sections and be represented as a bag of instances, multi-instance multi-label learning framework should fit this problem better. In this paper, we improve the performance of tag suggestion by using multi-instance multi-label learning. Each Web page is divided into a bag of instances. The experiments on real-word data from delicious suggest that our framework has better performance than traditional multi-label learning methods on the task of tag recommendation.

Along with the explosive growth of the Web, Web image search has become a more and more popular application which helps users digest the large amount of online visual information. Previous research mainly exploits visual information between images while rarely uses the text information surrounding the images on the Web pages. In this paper, we consider the relevance feedback as a machine learning problem. We proposed a novel relevance feedback framework for Web image search, which exploit both text and image modalities information with semi-supervised learning techniques. In each round of relevance feedbacks, the framework trains two classifiers for the two modalities by using the feedback information collected from the user. Then, it uses the unlabeled search result to improve these two classifiers. Finally, the ranked results list produced by image and text modality classifiers are combined to get the final rank. Experiments demonstrate the promise of the proposed framework.

Generating similar queries for a query, named query suggestion, is an important technology for helping search engine users. Since query data is very diverse and sparse, it is still challenging to measure the similarity of each query pair. We propose a novel algorithm called QueryTrans, which can efficiently compute pairwise similarity scores between all queries with respect to the global structure of a query trace graph mined from search engine logs. Compared with previous query suggestion approaches, QueryTrans is robust for different queries and stable for different parameter settings. We also present the performance of QueryTrans on large scale query logs. Experiments on about 100,000 queries show: QueryTrans can efficiently computes almost 10 billion pairwise similarity scores within 15 minutes on a single computer; and its results are significantly better than all 4 recent approaches on query suggestion.

Many real world applications increasingly involve both structured data and text, and entity based retrieval is an important problem in this realm. In this paper, we present an automatic query reformulation approach based on entities detected in each query. The aim is to utilize semantics associated with entities for enhancing document retrieval. This is done by expanding a query with terms/phrases related to entities in the query. We exploit Wikipedia as a large repository of entity information. Our reformulated approach consists of three major steps : (1) detect representative entity in a query; (2) expand the query with entity related terms/phrases; and (3) facilitate term dependency features. We evaluate our approach in ad-hoc retrieval task on four TREC collections, including two large web collections. Experiments results show that significant improvement is possible by utilizing entity corresponding information.

Previous work on term dependency has not taken into account semantic information underlying query phrases. In this work, we study the impact of utilizing phrase based concepts for term dependency. We use Wikipedia to separate important and less important term dependencies, and treat them accordingly as features in a linear feature-based retrieval model. We compare our method with a Markov Random Field (MRF) model on four TREC document collections. Our experimental results show that utilizing phrase based concepts improves the retrieval effectiveness of term dependency, and reduces the size of the feature set to large extent.

Previous on-line index maintenance strategies are mainly designed for document insertions without considering document deletions. In a truly dynamic search environment, however, documents may be added to and removed from the collection at any point in time. In this paper, we examine issues of on-line index maintenance with support for instantaneous document deletions and insertions. We present a DBT Merge strategy that can dynamically adjust the sequence of sub-index merge operations during index construction, and offers better query processing performance than previous methods, while providing an equivalent level of index maintenance performance when document insertions and deletions exist in parallel. Using experiments on 426 GB of web data we demonstrate the efficiency of our method in practice, showing that on-line index construction for dynamic text collections can be performed efficiently and almost as fast as for growing text collections.

In this paper we present a novel strategy, DragPushing, for improving the performance of text classifiers. The strategy is generic and takes advantage of training errors to successively refine the classification model of a base classifier. We describe how it is applied to generate two new classification algorithms; a Refined Centroid Classifier and a Refined Naïve Bayes Classifier. We present an extensive experimental evaluation of both algorithms on three English collections and one Chinese corpus. The results indicate that in each case, the refined classifiers achieve significant performance improvement over the base classifiers used. Furthermore, the performance of the Refined Centroid Classifier implemented is comparable, if not better, to that of state-of-the-art support vector machine (SVM)-based classifier, but offers a much lower computational cost.

We present a novel algorithm, DragPushing, for automatic text classification. Using a training data set, the algorithm first calculates the prototype vectors, or centroids, for each of the available document classes. Using misclassified examples, it then iteratively refines these centroids; by dragging the centroid of a correct class towards a misclassified example and in the same time pushing the centroid of an incorrect class away from the misclassified example. The algorithm is simple to implement and is computationally very efficient. Evaluation experiments conducted on two benchmark collections show that its classification accuracy is comparable to that of more complex methods, such as support vector machines (SVM).

This paper proposes an example-based phrase translation method in a Chinese to English cross-language information retrieval (CLIR) system. The method can generate much more accurate query translations than dictionary-based and common MT-based methods, and then improves the retrieval performance of our CLIR system.