Alignment plays a crucial role in Large Language Models (LLMs) in aligning with human preferences on a specific task/domain. Traditional alignment methods suffer from catastrophic forgetting, where models lose previously learned values when adapting to new preferences or domains. We introduce LifeAlign, a novel framework for lifelong alignment that enables LLMs to maintain consistent human preference alignment across sequential learning tasks without forgetting previously learned values. Our approach consists of two key innovations. First, we propose a focalized preference optimization strategy that aligns LLMs with new preferences while preventing the erosion of alignment acquired from previous tasks. Second, we develop a short-to-long memory consolidation mechanism that merges denoised short-term preference representations into stable long-term memory using intrinsic dimensionality reduction, enabling efficient storage and retrieval of alignment patterns across diverse domains. We evaluate LifeAlign across multiple sequential alignment tasks spanning different domains and preference types. Experimental results demonstrate that our method achieves superior performance in maintaining both preference alignment quality and knowledge retention compared to existing lifelong learning approaches.

Kedi Chen, Dezhao Ruan, Yuhao Dan, Yaoting Wang, Siyu Yan, Xuecheng Wu, Yinqi Zhang, Qin Chen, Jie Zhou, Liang He, Biqing Qi, Linyang Li, Qipeng Guo, Xiaoming Shi, Wei Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Knowledge editing is a promising method for updating Large Language Models efficiently.However, previous studies often suffer from poor specificity in continual editing, as they typically focus on single edits or preventing knowledge forgetting.To address this, we propose TamEdit, a trajectory-aware meta-learning method that preserves specificity for continual knowledge editing.TamEdit unifies three levels: Inner Optimization performs multi-step fast fine-tuning on the single edit; Trajectorybased Editing unifies continual edits with a growing memory; and Outer Optimization leverages meta-learning to distill cross-task strategies for preserving specificity.By capturing the relationships between different single edits within the trajectory, our method learns how to effectively avoid specificity drift.Experiments across multiple LLMs show TamEdit significantly outperforms baselines in continual editing, improving specificity by 14.81% with sub-second inference speed (0.55s per edit), while preserving general capabilities.

Multimodal large language models (MLLMs) have attracted considerable attention for their impressive capabilities in understanding and generating visual-language content, particularly in tasks such as visual question answering (VQA). However, the rapid evolution of knowledge in real-world applications poses challenges for these models: offline training becomes increasingly costly, and exposure to non-stationary data streams often leads to catastrophic forgetting. In this paper, we propose CL-MoE+, a dual-momentum Mixture-of-Experts (MoE) framework based on MLLMs for continual VQA. Our method integrates continual learning into MLLMs to leverage the rich commonsense knowledge embedded in large language models.We introduce a Dual-Router MoE (RMoE) module that selects both global and local experts through task-level and instance-level routers, enabling robust and context-aware expert allocation. Furthermore, we design an adaptive Momentum MoE (MMoE) to update experts’ parameters based on the knowledge drift degree and their relevance to specific tasks, thereby facilitating knowledge integration without forgetting. Extensive experiments on a 10-task split of the VQA v2 benchmark demonstrate that CL-MoE+ achieves state-of-the-art performance, validating its effectiveness in both retaining historical knowledge and learning new information in the continual learning setting.

Large Language Models (LLMs) are prone to hallucination with non-factual or unfaithful statements, which undermines the applications in real-world scenarios. Recent researches focus on uncertainty-based hallucination detection, which utilizes the output probability of LLMs for uncertainty calculation and does not rely on external knowledge or frequent sampling from LLMs. Whereas, most approaches merely consider the uncertainty of each independent token, while the intricate semantic relations among tokens and sentences are not well studied, which limits the detection of hallucination that spans over multiple tokens and sentences in the passage. In this paper, we propose a method to enhance uncertainty modeling with semantic graph for hallucination detection. Specifically, we first construct a semantic graph that well captures the relations among entity tokens and sentences. Then, we incorporate the relations between two entities for uncertainty propagation to enhance sentence-level hallucination detection. Given that hallucination occurs due to the conflict between sentences, we further present a graph-based uncertainty calibration method that integrates the contradiction probability of the sentence with its neighbors in the semantic graph for uncertainty calculation. Extensive experiments on two datasets show the great advantages of our proposed approach. In particular, we obtain substantial improvements with 19.78% in passage-level hallucination detection.

Retrieval-augmented generation (RAG) is usually integrated into large language models (LLMs) to mitigate hallucinations and knowledge obsolescence.Whereas, conventional onestep retrieve-and-read methods are insufficient for multi-hop question answering, facing challenges of retrieval semantic mismatching and the high cost in handling interdependent subquestions.In this paper, we propose Optimizing Question Semantic Space for Dynamic Retrieval-Augmented Multi-hop Question Answering (Q-DREAM).Q-DREAM consists of three key modules: (1) the Question Decomposition Module (QDM), which decomposes multi-hop questions into fine-grained subquestions; (2) the Subquestion Dependency Optimizer Module (SDOM), which models the interdependent relations of subquestions for better understanding; and (3) the Dynamic Passage Retrieval Module (DPRM), which aligns subquestions with relevant passages by optimizing the semantic embeddings.Experimental results across various benchmarks demonstrate that Q-DREAM significantly outperforms existing RAG methods, achieving state-of-the-art performance in both in-domain and out-of-domain settings.Notably, Q-DREAM also improves retrieval efficiency while maintaining high accuracy compared with recent baselines.

Multimodal large language models (MLLMs) have garnered widespread attention from researchers due to their remarkable understanding and generation capabilities in visual language tasks (e.g., visual question answering). However, the rapid pace of knowledge updates in the real world makes offline training of MLLMs costly, and when faced with non-stationary data streams, MLLMs suffer from catastrophic forgetting during learning. In this paper, we propose an MLLMs-based dual momentum Mixture-of-Experts (CL-MoE) framework for continual visual question answering (VQA). We integrate MLLMs with continual learning to utilize the rich commonsense knowledge in LLMs. We introduce a Dual-Router MoE (RMoE) strategy to select the global and local experts using task-level and instance-level routers, to robustly assign weights to the experts most appropriate for the task. Then, we design a dynamic Momentum MoE (MMoE) to update the parameters of experts dynamically based on the relationships between the experts and tasks/instances, so that the model can absorb new knowledge while maintaining existing knowledge. The extensive experimental results indicate that our method achieves state-of-the-art performance on 10 VQA tasks, proving the effectiveness of our approach.

Automatic math correction aims to check students’ solutions to mathematical problems via artificial intelligence technologies. Most existing studies focus on judging the final answer at the problem level, while they ignore detailed feedback on each step in a math problem-solving process, which requires abilities of semantic understanding and reasoning. In this paper, we propose a reinforcement learning (RL)-based method to boost large language model (LLM) for step-level automatic math correction, named StepAMC. Particularly, we convert the step-level automatic math correction within the text classification task into an RL problem to enhance the reasoning capabilities of LLMs. Then, we design a space-constrained policy network to improve the stability of RL. Then, we introduce a fine-grained reward network to convert the binary human feedback into a continuous value. We conduct extensive experiments over two benchmark datasets and the results show that our model outperforms the eleven strong baselines.

Large language models (LLMs) have obtained promising results in mathematical reasoning, a foundational human intelligence skill. Most previous studies focus on improving or measuring the performance of LLMs via textual math datasets (e.g., MATH, GSM8K). In this paper, we release a Chinese multimodal math (CMM-Math) dataset, including benchmark and training parts, to evaluate and enhance the mathematical reasoning of LMMs. CMM-Math contains over 28,000 high-quality samples, featuring a variety of problem types (e.g., choice, fill-in-the-blank, analysis) with detailed solutions across 12 grade levels from elementary to high school in China. The problem may contain multiple images, and the visual context may be present in the questions or opinions, which makes this dataset more challenging. Our comprehensive analysis reveals that state-of-the-art LMMs on the CMM-Math face challenges, emphasizing the necessity for further improvements in LMM development. We also propose a Multimodal Mathematical LMM (Math-LMM) to handle the problems with mixed input of multiple images and text segments. The Math-LMM is trained using three stages: foundational pre-training, foundational fine-tuning, and mathematical fine-tuning. The extensive experiments indicate that our model effectively improves math reasoning performance by comparing it with the SOTA LMMs over three multimodal mathematical datasets. We release the datasets on GitHub (https://github.com/ECNU-ICALK/EduChat-Math) and Huggingface (https://huggingface.co/datasets/ecnu-icalk/cmm-math).

Large language models (LLMs) have shown great success in various Natural Language Processing (NLP) tasks, whist they still need updates after deployment to fix errors or keep pace with the changing knowledge in the world. Researchers formulate such problem as Model Editing and have developed various editors focusing on different axes of editing properties. However, current editors can hardly support all properties and rely on heavy computational resources. In this paper, we propose a plug-in Model Editing method based on neuron-indexed dynamic LoRA (MELO), which alters the behavior of language models by dynamically activating certain LoRA blocks according to the index built in an inner vector database. Our method satisfies various editing properties with high efficiency and can be easily integrated into multiple LLM backbones. Experimental results show that our proposed MELO achieves state-of-the-art editing performance on three sequential editing tasks (document classification, question answering and hallucination correction), while requires the least trainable parameters and computational cost.

With the introduction of large language models (LLMs), automatic math reasoning has seen tremendous success. However, current methods primarily focus on providing solutions or using techniques like Chain-of-Thought to enhance problem-solving accuracy. In this paper, we focus on improving the capability of mathematics teaching via a Socratic teaching-based LLM (SocraticLLM), which guides learners toward profound thinking with clarity and self-discovery via conversation. We collect and release a high-quality mathematical teaching dataset, named SocraticMATH, which provides Socratic-style conversations of problems with extra knowledge. Also, we propose a knowledge-enhanced LLM as a strong baseline to generate reliable responses with review, guidance/heuristic, rectification, and summarization. Experimental results show the great advantages of SocraticLLM by comparing it with several strong generative models. The codes and datasets are available on https://github.com/ECNU-ICALK/SocraticMath.

Aspect-Based Sentiment Analysis (ABSA) stands as a crucial task in predicting the sentiment polarity associated with identified aspects within text. However, a notable challenge in ABSA lies in precisely determining the aspects’ boundaries (start and end indices), especially for long ones, due to users’ colloquial expressions. We propose DiffusionABSA, a novel diffusion model tailored for ABSA, which extracts the aspects progressively step by step. Particularly, DiffusionABSA gradually adds noise to the aspect terms in the training process, subsequently learning a denoising process that progressively restores these terms in a reverse manner. To estimate the boundaries, we design a denoising neural network enhanced by a syntax-aware temporal attention mechanism to chronologically capture the interplay between aspects and surrounding text. Empirical evaluations conducted on eight benchmark datasets underscore the compelling advantages offered by DiffusionABSA when compared against robust baseline models. Our code is publicly available at https://github.com/Qlb6x/DiffusionABSA.

Information extraction (IE) aims to extract complex structured information from the text. Numerous datasets have been constructed for various IE tasks, leading to time-consuming and labor-intensive data annotations. Nevertheless, most prevailing methods focus on training task-specific models, while the common knowledge among different IE tasks is not explicitly modeled. Moreover, the same phrase may have inconsistent labels in different tasks, which poses a big challenge for knowledge transfer using a unified model. In this study, we propose a regularization-based transfer learning method for IE (TIE) via an instructed graph decoder. Specifically, we first construct an instruction pool for datasets from all well-known IE tasks, and then present an instructed graph decoder, which decodes various complex structures into a graph uniformly based on corresponding instructions. In this way, the common knowledge shared with existing datasets can be learned and transferred to a new dataset with new labels. Furthermore, to alleviate the label inconsistency problem among various IE tasks, we introduce a task-specific regularization strategy, which does not update the gradients of two tasks with ‘opposite direction’. We conduct extensive experiments on 12 datasets spanning four IE tasks, and the results demonstrate the great advantages of our proposed method.

Gradient-based explanation methods are increasingly used to interpret neural models in natural language processing (NLP) due to their high fidelity. Such methods determine word-level importance using dimension-level gradient values through a norm function, often presuming equal significance for all gradient dimensions. However, in the context of Aspect-based Sentiment Analysis (ABSA), our preliminary research suggests that only specific dimensions are pertinent. To address this, we propose the Information Bottleneck-based Gradient (IBG) explanation framework for ABSA. This framework leverages an information bottleneck to refine word embeddings into a concise intrinsic dimension, maintaining essential features and omitting unrelated information. Comprehensive tests show that our IBG approach considerably improves both the models’ performance and the explanations’ clarity by identifying sentiment-aware features.

Domain adaption has been widely adapted for cross-domain sentiment analysis to transfer knowledge from the source domain to the target domain. Whereas, most methods are proposed under the assumption that the target (test) domain is known, making them fail to generalize well on unknown test data that is not always available in practice. In this paper, we focus on the problem of domain generalization for cross-domain sentiment analysis. Specifically, we propose a backdoor adjustment-based causal model to disentangle the domain-specific and domain-invariant representations that play essential roles in tackling domain shift. First, we rethink the cross-domain sentiment analysis task in a causal view to model the causal-and-effect relationships among different variables. Then, to learn an invariant feature representation, we remove the effect of domain confounders (e.g., domain knowledge) using the backdoor adjustment. A series of experiments over many homologous and diverse datasets show the great performance and robustness of our model by comparing it with the state-of-the-art domain generalization baselines.

Document-level Event Causality Identification (DECI) aims to identify causal relations between two events in documents. Recent researches tend to use pre-trained language models to generate the event causal relations. Whereas, these methods are prone to the errors of sequential generation due to multiple events in a document. Moreover, the potential structures such as event coreference and related causal chain are neglected. In this paper, we propose a multi-task learning framework to enhance event causality identification with rationale and structure-aware causal question answering. Specifically, the DECI task is transformed into multiple-choice question answering, and the causes and effects of the questioned event are generated with large language models. In addition, we generate the rationales to explain why these events have causal relations. Moreover, we construct an event structure graph, which models the multi-hop potential relations for causal reasoning of the current event. Experiments on two benchmark datasets show the great advantages of our proposed approach compared to the state-of-the-art methods. Moreover, we conduct both quantitative and qualitative analyses, which shed light on why each component of our approach can lead to great improvements.

Retrieval-Augmented Generation (RAG) aims to generate more reliable and accurate responses, by augmenting large language models(LLMs) with the external vast and dynamic knowledge. Most previous work focuses on using RAG for single-round question answering, while how to adapt RAG to the complex conversational setting wherein the question is interdependent on the preceding context is not well studied. In this paper, we propose a conversation-level RAG (ConvRAG) approach, which incorporates fine-grained retrieval augmentation and self-check for conversational question answering (CQA). In particular, our approach consists of three components, namely conversational question refiner, fine-grained retriever and self-check based response generator, which work collaboratively for question understanding and relevant information acquisition in conversational settings. Extensive experiments demonstrate the great advantages of our approach over the state-of-the-art baselines. Moreover, we also release a Chinese CQA dataset with new features including reformulated question, extracted keyword, retrieved paragraphs and their helpfulness, which facilitates further researches in RAG enhanced CQA.

Federated Learning (FL) has drawn much attention for distributed system over the Internet of Things (IoT), since it enables collaborative machine learning on heterogeneous devices while resolves concerns about privacy leakage. Due to the catastrophic forgetting (CF) phenomenon of optimization methods, existing FL approaches are restricted to single task learning and typically assume that data from all nodes are simultaneously available during training. However, in practical IoT scenarios, the data preparation from nodes may be asynchronous, and different tasks require incremental training. To address the issues, we propose a continual FL (CFL) framework with parameter-level importance aggregation (PI-Fed), which supports collaborative task-incremental learning with privacy preservation. Specifically, PI-Fed evaluates the importance of each parameter in the global model to all history tasks, which is computed locally and aggregated at the center server. Then the server performs soft-masking on the averaged gradient collected from local clients based on the parameter importance. By minimizing the change on important parameters, PI-Fed effectively overcomes CF and also achieves high efficiency without experience replay. Extensive experiments on 4 benchmarks with at most 20 sequential tasks demonstrate that our proposed PI-Fed significantly outperforms traditional FL baselines (FedAvg, FedNova, and SCAFFOLD).

As the primary subtask of sentiment analysis, aspect-based sentiment classification (ABSC) aims to predict the sentiment polarity for a given aspect. While recent deep neural models for ABSC have shown good performance, their robustness is limited due to their reliance on spurious correlations between aspects and sentiment. Specifically, most existing models tend to assign the most frequent sentiment label to a certain aspect in different sentences, or assume different aspects in the same sentence to have the same sentiment polarity, which is vulnerable due to the aspect biases. In this paper, we propose a causal graph to identify and analyze the causal relationships among treatment variables (e.g., aspect, sentence), intermediate variables (e.g., aspect-aware content), and outcome variables (e.g., sentiment polarity) for ABSC. To address the issue of spurious relationships that mislead sentiment polarity prediction, we introduce a novel causal inference framework called CausalABSC. CausalABSC is model agnostic, allowing integration into existing methods. We conduct extensive experiments on five benchmark datasets, which demonstrate the state-of-the-art performance of CausalABSC and its effectiveness in debiasing.

LiDAR-camera fusion methods have shown impressive performance in 3D object detection. Recent advanced multi-modal methods mainly perform global fusion, where image features and point cloud features are fused across the whole scene. Such practice lacks fine-grained region-level information, yielding suboptimal fusion performance. In this paper, we present the novel Local-to-Global fusion network (LoGoNet), which performs LiDAR-camerafusion at both local and global levels. Concretely, the Global Fusion (GoF) of LoGoNet is built upon previous literature, while we exclusively use point centroids to more precisely represent the position of voxel features, thus achieving better crossmodal alignment. As to the Local Fusion (LoF), we first divide each proposal into uniform grids and then project these grid centers to the images. The image features around the projected grid points are sampled to be fused with position-decorated point cloud features, maximally uti-lizing the rich contextual information around the proposals. The Feature Dynamic Aggregation (FDA) module is further proposed to achieve information interaction between these locally and globally fused features, thus producing more informative multi-modal features. Extensive experiments on both Waymo Open Dataset (WOD) and KITTI datasets show that LoGoNet outperforms all state-of-the-art 3D detection methods. Notably, LoGoNet ranks 1st on Waymo 3D object detection leaderboard and obtains 81.02 mAPH (L2) detection performance. It is noteworthy that, for the first time, the detection performance on three classes surpasses 80 APH (L2) simultaneously. Code will be available at https://github.com/sankin97/LoGoNet.

Gradient-based explanation methods play an important role in the field of interpreting complex deep neural networks for NLP models. However, the existing work has shown that the gradients of a model are unstable and easily manipulable, which impacts the model’s reliability largely. According to our preliminary analyses, we also find the interpretability of gradient-based methods is limited for complex tasks, such as aspect-based sentiment classification (ABSC). In this paper, we propose an Interpretation-Enhanced Gradient-based framework for ABSC via a small number of explanation annotations, namely IGA. Particularly, we first calculate the word-level saliency map based on gradients to measure the importance of the words in the sentence towards the given aspect. Then, we design a gradient correction module to enhance the model’s attention on the correct parts (e.g., opinion words). Our model is model agnostic and task agnostic so that it can be integrated into the existing ABSC methods or other tasks. Comprehensive experimental results on four benchmark datasets show that our IEGA can improve not only the interpretability of the model but also the performance and robustness.

Visual dialog aims to address a sequence of questions by effectively reasoning over both the dialog history and image content. While existing methods primarily focus on devising various attention mechanisms to capture interactions between different modalities, explicit signals encouraging semantic alignment in the visual dialog are seldom utilized. In this paper, we present a novel approach that leverages Optimal Transport to provide explicit and interpretable training signals to guide intra- and inter-modal alignment for the text and image in the visual dialog. Specifically, our approach consists of two kinds of alignment modules, Word-Word Alignment (WWA) and Region-Word Alignment (RWA). The WWA module learns latent relationships between a given question and a dialog history to align different concepts or pronouns that represent the same entity. As for the RWA module, it models the internal structures of text and images with graphs and performs graph matching for region-word alignment. We perform experiments on the benchmark dataset Visdial v1.0, and the experimental results show that our proposed approach achieves new state-of-the-art performance with respect to most metrics.

In the scenario of Model-as-a-Service (MaaS), pre-trained models are usually released as inference APIs. Users are allowed to query those models with manually crafted prompts. Without accessing the network structure and gradient information, it's tricky to perform continuous prompt tuning on MaaS, especially for vision-language models (VLMs) considering cross-modal interaction. In this paper, we propose a black-box prompt tuning framework for VLMs to learn task-relevant prompts without back-propagation. In particular, the vision and language prompts are jointly optimized in the intrinsic parameter subspace with various evolution strategies. Different prompt variants are also explored to enhance the cross-model interaction. Experimental results show that our proposed black-box prompt tuning framework outperforms both hand-crafted prompt engineering and gradient-based prompt learning methods, which serves as evidence of its capability to train task-relevant prompts in a derivative-free manner.

Event argument extraction (EAE) aims to extract arguments with given roles from texts, which have been widely studied in natural language processing. Most previous works have achieved good performance in specific EAE datasets with dedicated neural architectures. Whereas, these architectures are usually difficult to adapt to new datasets/scenarios with various annotation schemas or formats. Furthermore, they rely on large-scale labeled data for training, which is unavailable due to the high labelling cost in most cases. In this paper, we propose a multi-format transfer learning model with variational information bottleneck, which makes use of the information especially the common knowledge in existing datasets for EAE in new datasets. Specifically, we introduce a shared-specific prompt framework to learn both format-shared and format-specific knowledge from datasets with different formats. In order to further absorb the common knowledge for EAE and eliminate the irrelevant noise, we integrate variational information bottleneck into our architecture to refine the shared representation. We conduct extensive experiments on three benchmark datasets, and obtain new state-of-the-art performance on EAE.

Zero-shot text classification (ZSTC) poses a big challenge due to the lack of labeled data for unseen classes during training. Most studies focus on transferring knowledge from seen classes to unseen classes, which have achieved good performance in most cases. Whereas, it is difficult to transfer knowledge when the classes have semantic gaps or low similarities. In this paper, we propose a prompt-based method, which enhances semantic understanding for each class and learns the matching between texts and classes for better ZSTC. Specifically, we first generate discriminative words for class description with prompt inserting (PIN). Then, a prompt matching (POM) model is learned to determine whether the text can well match the class description. Experiments on three benchmark datasets show the great advantages of our proposed method. In particular, we achieve the state-of-the-art performance on the unseen classes, while maintaining comparable strength with the existing ZSTC approaches regarding to the seen classes.

Implicit event argument extraction (EAE) aims to identify arguments that could scatter over the document. Most previous work focuses on learning the direct relations between arguments and the given trigger, while the implicit relations with long-range dependency are not well studied. Moreover, recent neural network based approaches rely on a large amount of labeled data for training, which is unavailable due to the high labelling cost. In this paper, we propose a Curriculum learning based Prompt tuning (CUP) approach, which resolves implicit EAE by four learning stages. The stages are defined according to the relations with the trigger node in a semantic graph, which well captures the long-range dependency between arguments and the trigger. In addition, we integrate a prompt-based encoder-decoder model to elicit related knowledge from pre-trained language models (PLMs) in each stage, where the prompt templates are adapted with the learning progress to enhance the reasoning for arguments. Experimental results on two well-known benchmark datasets show the great advantages of our proposed approach. In particular, we outperform the state-of-the-art models in both fully-supervised and low-data scenarios.

Structured sentiment analysis, which aims to extract the complex semantic structures such as holders, expressions, targets, and polarities, has obtained widespread attention from both industry and academia. Unfortunately, the existing structured sentiment analysis datasets refer to a few languages and are relatively small, limiting neural network models' performance. In this paper, we focus on the cross-lingual structured sentiment analysis task, which aims to transfer the knowledge from the source language to the target one. Notably, we propose a Knowledge-Enhanced Adversarial Model (KEAM) with both implicit distributed and explicit structural knowledge to enhance the cross-lingual transfer. First, we design an adversarial embedding adapter for learning an informative and robust representation by capturing implicit semantic information from diverse multi-lingual embeddings adaptively. Then, we propose a syntax GCN encoder to transfer the explicit semantic information (e.g., universal dependency tree) among multiple languages. We conduct experiments on five datasets and compare KEAM with both the supervised and unsupervised methods. The extensive experimental results show that our KEAM model outperforms all the unsupervised baselines in various metrics.

Previous studies about event-level sentiment analysis (SA) usually model the event as a topic, a category or target terms, while the structured arguments (e.g., subject, object, time and location) that have potential effects on the sentiment are not well studied. In this paper, we redefine the task as structured event-level SA and propose an End-to-End Event-level Sentiment Analysis (E3SA) approach to solve this issue. Specifically, we explicitly extract and model the event structure information for enhancing event-level SA. Extensive experiments demonstrate the great advantages of our proposed approach over the state-of-the-art methods. Noting the lack of the dataset, we also release a large-scale real-world dataset with event arguments and sentiment labelling for promoting more researches.

Users usually consult the manufacturers or the internet when they encounter operation questions with an electronics product. In this paper, we explore to represent an operation question as a procedure graph and formulate the problem of operation diagnosis as two sub-tasks, namely error node detection, and correction, on top of the graph. We construct the first benchmark for this task and propose a transformer-based model to integrate external knowledge and context information to enhance the performance. Experimental results show the effectiveness of our proposed model.

Visual grounding has attracted much attention with the popularity of vision language. Existing one-stage methods are far ahead of two-stage methods in speed. However, these methods fuse the textual feature and visual feature map by simply concatenation, which ignores the textual semantics and limits these models' ability in cross-modal understanding. To overcome this weakness, we propose a semantic-aware framework that utilizes both queries' structured knowledge and context-sensitive representations to filter the visual feature maps to localize the referents more accurately. Our framework contains an entity filter, an attribute filter, and a location filter. These three filters filter the input visual feature map step by step according to each query's aspects respectively. A grounding module further regresses the bounding boxes to localize the referential object. Experiments on various commonly used datasets show that our framework achieves a real-time inference speed and outperforms all state-of-the-art methods.

Knowledge of a disease includes information of various aspects of the disease, such as signs and symptoms, diagnosis and treatment.This disease knowledge is critical for many healthrelated and biomedical tasks, including consumer health question answering, medical language inference and disease name recognition.While pre-trained language models like BERT have shown success in capturing syntactic, semantic, and world knowledge from text, we find they can be further complemented by specific information like knowledge of symptoms, diagnoses, treatments, and other disease aspects.Hence, we integrate BERT with disease knowledge for improving these important tasks.Specifically, we propose a new disease knowledge infusion training procedure and evaluate it on a suite of BERT models including BERT, BioBERT, SciBERT, Clinical-BERT, BlueBERT, and ALBERT.Experiments over the three tasks show that these models can be enhanced in nearly all cases, demonstrating the viability of disease knowledge infusion.For example, accuracy of BioBERT on consumer health question answering is improved from 68.29% to 72.09%, while new SOTA results are observed in two datasets.We make our data and code freely available.

This paper focuses on a novel financial event prediction task that takes a historical event chain as input and predicts what event will happen next. We introduce financial news as supplementary information to solve problems of multiple interpretations of same financial event. Besides, a gated graph neural network based approach is utilized to capture complicated relationships between event graphs for better event prediction. For the evaluation, we build a new dataset consisting of financial events for thousands of Chinese listed companies from 2013 to 2017. Experimental results show the effectiveness of our proposed model.

Xinzhu Lin, Xiahui He, Qin Chen, Huaixiao Tou, Zhongyu Wei, Ting Chen. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

The recurrent neural networks (RNNs) have shown good performance for sentence similarity modeling in recent years. Most RNNs focus on modeling the hidden states based on the current sentence, while the context information from the other sentence is not well investigated during the hidden state generation. In this paper, we propose a context-aligned RNN (CA-RNN) model, which incorporates the contextual information of the aligned words in a sentence pair for the inner hidden state generation. Specifically, we first perform word alignment detection to identify the aligned words in the two sentences. Then, we present a context alignment gating mechanism and embed it into our model to automatically absorb the aligned words' context for the hidden state update. Experiments on three benchmark datasets, namely TREC-QA and WikiQA for answer selection and MSRP for paraphrase identification, show the great advantages of our proposed model. In particular, we achieve the new state-of-the-art performance on TREC-QA and WikiQA. Furthermore, our model is comparable to if not better than the recent neural network based approaches on MSRP.

We mainly investigate word influence in neural sentiment classification, which results in a novel approach to promoting word sentiment and negation as attentions. Particularly, a sentiment and negation neural network (SNNN) is proposed, including a sentiment neural network (SNN) and a negation neural network (NNN). First, we modify the word level by embedding the word sentiment and negation information as the extra layers for the input. Second, we adopt a hierarchical LSTM model to generate the word-level, sentence-level and document-level representations respectively. After that, we enhance word sentiment and negation as attentions over the semantic level. Finally, the experiments conducting on the IMDB and Yelp data sets show that our approach is superior to the state-of-the-art baselines. Furthermore, we draw the interesting conclusions that (1) LSTM performs better than CNN and RNN for neural sentiment classification; (2) word sentiment and negation are a strong alliance with attention, while overfitting occurs when they are simultaneously applied at the embedding layer; and (3) word sentiment/negation can be singly implemented for better performance as both embedding layer and attention at the same time.

The neural networks have attracted great attention for sentence similarity modeling in recent years. Most neural networks focus on the representation of each sentence, while the common features of a sentence pair are not well studied. In this paper, we propose a Collaborative and Adversarial Network (CAN), which explicitly models the common features between two sentences for enhancing sentence similarity modeling. To be specific, a common feature extractor is presented and embedded into our CAN model, which includes a generator and a discriminator playing a collaborative and adversarial game for common feature extraction. Experiments on three benchmark datasets, namely TREC-QA and WikiQA for answer selection and MSRP for paraphrase identification, show that our proposed model is effective to boost the performance of sentence similarity modeling. In particular, our proposed model outperforms the state-of-the-art approaches on TREC-QA without using any external resources or pre-training. For the other two datasets, our model is also comparable to if not better than the recent neural network approaches.

Query expansion under the pseudo-relevance feedback (PRF) framework has been extensively studied in information retrieval. However, most expansion methods are mainly based on the statistics of single terms, which can generate plenty of irrelevant query terms and decrease retrieval performance. To alleviate this problem, we propose an approach that adapts the PRF-based contextual snippets into a context-aware topic model to enhance query representations. Specifically, instead of selecting a series of independent terms, we make full use of the query contextual information and focus on the snippets with the length ofnin the PRF documents. Furthermore, we propose a context-aware topic (CAT) model to mine the topic distributions of the query-relevant snippets, namely, fine contextual snippets. In contrast to the traditional topic models that infer the topics from the whole corpus, we establish a bridge between the snippets and the corresponding PRF documents, which can be used for modeling the topics more precisely and efficiently. Finally, the topic distributions of the fine snippets are used for context-aware and topic-sensitive query representations. To evaluate the performance of our approach, we integrate the obtained queries into a topic-based hybrid retrieval model and conduct extensive experiments on various TREC collections. The experimental results show that our query-modeling approach is more effective in boosting retrieval performance compared with the state-of-the-art methods.

Temporal information has been widely used to promote the information retrieval (IR) performance, especially for microblog search which usually prefers the latest news and events. Previous studies mainly focused on incorporating the document-level temporal information into retrieval, while the temporal relevance of each query word was not well investigated. In this paper, we propose a word temporal predictor to characterize the word-level temporal relevance by fine-grained time-aware kernel density estimation over the feedback documents. In addition, we present a fine-grained time-aware framework to integrate the proposed word temporal predictor with the traditional document temporal predictor for retrieval. Finally, we incorporate the framework into two state-of-the-art retrieval models, namely language model (LM) and BM25. The experimental results on the TREC 2011-2014 Microblog collections, show that our proposed word temporal predictor is effective to boost the retrieval performance within both LM and BM25 frameworks. In particular, we achieve significant improvements over the strong baselines with optimized settings in most cases. Furthermore, our fine-grained time-aware models with word temporal predictor are comparable to if not better than the state-of-the-art temporal retrieval models.

In recommender systems, the essential task is to predict the personalized rating of a user to a new item. To address this task, recommender systems usually employ matrix factorization model to predict ratings over a user-item rating matrix. However, this model severely suffers from the problem of data sparsity. Noting the large amount of user reviews available in the Internet, we exploit user preferences and item attributes contained in reviews to alleviate the problem. Specifically, we propose a Neural Profile-Aware Matrix Factorization model, namely NPMF, which incorporates the user and item profiles modeled with neural networks for rating prediction. We evaluate the performance of NPMF using three large-scale real-world datasets released in Yelp. The experimental results show that NPMF outperforms other mainstream rating prediction techniques and indeed alleviates the data sparsity problem.

Attention based recurrent neural networks (RNN) have shown a great success for question answering (QA) in recent years. Although significant improvements have been achieved over the non-attentive models, the position information is not well studied within the attention-based framework. Motivated by the effectiveness of using the word positional context to enhance information retrieval, we assume that if a word in the question (i.e., question word) occurs in an answer sentence, the neighboring words should be given more attention since they intuitively contain more valuable information for question answering than those far away. Based on this assumption, we propose a positional attention based RNN model, which incorporates the positional context of the question words into the answers' attentive representations. Experiments on two benchmark datasets show the great advantages of our proposed model. Specifically, we achieve a maximum improvement of 8.83% over the classical attention based RNN model in terms of mean average precision. Furthermore, our model is comparable to if not better than the state-of-the-art approaches for question answering.

Here we propose an advance Skip-gram model to incorporate both word sentiment and negation information. In particular, there is a a softmax layer for the word sentiment polarity upon the Skip-gram model. Then, two paralleled embedding layers are set up in the same embedding space, one for the affirmative context and the other for the negated context, followed by their loss functions. We evaluate our proposed model on the 2013 and 2014 SemEval data sets. The experimental results show that the proposed approach achieves better performance and learns higher dimensional word embedding informatively on the large-scale data.