As Large Language Models (LLMs) continue to advance in capability and influence, ensuring their security and preventing harmful outputs has become crucial. A promising approach to address these concerns involves training models to automatically generate adversarial prompts for red teaming. However, the evolving subtlety of vulnerabilities in LLMs challenges the effectiveness of current adversarial methods, which struggle to generate diverse, complex prompts and dynamically explore the weaknesses of these models. To tackle these challenges, we introduce the Self-Evolving Adversarial Safety (SEAS) optimization framework, which includes both a SEAS dataset and a SEAS pipeline. The SEAS dataset comprises complex adversarial prompts, while the SEAS pipeline operates through three stages: Initialization, Attack, and Adversarial Optimization. This framework generates a diverse range of adversarial prompts and dynamically explores the model's vulnerabilities to enhance its security. Our contributions include a novel adversarial framework, a comprehensive safety dataset, and empirical evidence demonstrating the effectiveness of SEAS.

Large language models (LLMs) with one or more fine-tuning phases have become necessary to unlock various capabilities, enabling LLMs to follow natural language instructions and align with human preferences. However, it carries the risk of catastrophic forgetting during sequential training, the parametric knowledge or the ability learned in previous stages may be overwhelmed by incoming training data. This paper finds that LLMs can restore some original knowledge by regularly resetting partial parameters. Inspired by this, we introduce Half Fine-Tuning (HFT) for LLMs, as a substitute for full fine-tuning (FFT), to mitigate the forgetting issues, where half of the parameters are selected to learn new tasks. In contrast, the other half are frozen to retain previous knowledge. We provide a feasibility analysis from the optimization perspective and interpret the parameter selection operation as a regularization term. HFT could be seamlessly integrated into existing fine-tuning frameworks without changing the model architecture. Extensive experiments and analysis on supervised fine-tuning, direct preference optimization, and continual learning consistently demonstrate the effectiveness, robustness, and efficiency of HFT. Compared with FFT, HFT not only significantly alleviates the forgetting problem, but also achieves the best performance in a series of downstream benchmarks, with an approximately 30% reduction in training time.

Addressing the disparity between predictions and actual results can enable individuals to expand their thought processes and stimulate self-reflection, thus promoting accurate planning. In this research, we present PreAct, an agent framework that integrates prediction, reasoning, and action. By utilizing the information derived from predictions, the large language model (LLM) agent can provide a wider range and more strategically focused reasoning. This leads to more efficient actions that aid the agent in accomplishing intricate tasks. Our experimental results show that PreAct surpasses the ReAct method in completing complex tasks and that PreAct’s performance can be further improved when paired with other memory or selection strategy techniques. We presented the model with varying quantities of historical predictions and discovered that these predictions consistently enhance LLM planning. The variances in single-step reasoning between PreAct and ReAct indicate that PreAct indeed has benefits in terms of diversity and strategic orientation over ReAct.

Large Language Model (LLM) based agents have proved their ability to perform complex tasks like humans. However, there is still a large gap between open-sourced LLMs and commercial models like the GPT series. In this paper, we focus on improving the agent generalization capabilities of LLMs via instruction tuning. We first observe that the existing agent training corpus exhibits satisfactory results on held-in evaluation sets but fails to generalize to held-out sets. These agent-tuning works face severe formatting errors and are frequently stuck in the same mistake for a long while. We analyze that the poor generalization ability comes from overfitting to several manual agent environments and a lack of adaptation to new situations. They struggle with the wrong action steps and can not learn from the experience but just memorize existing observation-action relations. Inspired by the insight, we propose a novel AgentRefine framework for agent-tuning. The core idea is to enable the model to learn to correct its mistakes via observation in the trajectory. Specifically, we propose an agent synthesis framework to encompass a diverse array of environments and tasks and prompt a strong LLM to refine its error action according to the environment feedback. AgentRefine significantly outperforms state-of-the-art agent-tuning work in terms of generalization ability on diverse agent tasks. It also has better robustness facing perturbation and can generate diversified thought in inference. Our findings establish the correlation between agent generalization and self-refinement and provide a new paradigm for future research.

Large language models (LLMs) have demonstrated significant potential in advancing various fields of research and society. However, the current community of LLMs overly focuses on benchmarks for analyzing specific foundational skills (e.g. mathematics and code generation), neglecting an all-round evaluation of the computer science field. To bridge this gap, we introduce CS-Bench, the first multilingual (English, Chinese, French, German) benchmark dedicated to evaluating the performance of LLMs in computer science. CS-Bench comprises approximately 10K meticulously curated test samples, covering 26 subfields across 4 key areas of computer science, encompassing various task forms and divisions of knowledge and reasoning. Utilizing CS-Bench, we conduct a comprehensive evaluation of over 30 mainstream LLMs, revealing the relationship between CS performance and model scales. We also quantitatively analyze the reasons for failures in existing LLMs and highlight directions for improvements, including knowledge supplementation and CS-specific reasoning. Further cross-capability experiments show a high correlation between LLMs' capabilities in computer science and their abilities in mathematics and coding. Moreover, expert LLMs specialized in mathematics and coding also demonstrate strong performances in several CS subfields. Looking ahead, we envision CS-Bench serving as a cornerstone for LLM applications in the CS field and paving new avenues in assessing LLMs' diverse reasoning capabilities. Our project homepage is available at https://csbench.github.io/.

Faces and humans are crucial elements in social interaction and are widely included in everyday photos and videos. Therefore, a deep understanding of faces and humans will enable multi-modal assistants to achieve improved response quality and broadened application scope. Currently, the multi-modal assistant community lacks a comprehensive and scientific evaluation of face and human understanding abilities. In this paper, we first propose a hierarchical ability taxonomy that includes three levels of abilities. Then, based on this taxonomy, we collect images and annotations from publicly available datasets in the face and human community and build a semi-automatic data pipeline to produce problems for the new benchmark. Finally, the obtained Face-Human-Bench includes a development set and a test set, each with 1800 problems, supporting both English and Chinese. We conduct evaluations over 25 mainstream multi-modal large language models (MLLMs) with our Face-Human-Bench, focusing on the correlation between abilities, the impact of the relative position of targets on performance, and the impact of Chain of Thought (CoT) prompting on performance. We also explore which abilities of MLLMs need to be supplemented by specialist models. The dataset and evaluation code have been made publicly available at https://face-human-bench.github.io.

Long-horizon tasks requiring multi-step reasoning and dynamic re-planning remain challenging for large language models (LLMs). Sequential prompting methods are prone to context drift, loss of goal information, and recurrent failure cycles, while hierarchical prompting methods often weaken cross-level continuity or incur substantial runtime overhead. We introduce ReCAP (Recursive Context-Aware Reasoning and Planning), a hierarchical framework with shared context for reasoning and planning in LLMs. ReCAP combines three key mechanisms: (i) plan-ahead decomposition, in which the model generates a full subtask list, executes the first item, and refines the remainder; (ii) structured re-injection of parent plans, maintaining consistent multi-level context during recursive return; and (iii) memory-efficient execution, bounding the active prompt so costs scale linearly with task depth. Together these mechanisms align high-level goals with low-level actions, reduce redundant prompting, and preserve coherent context updates across recursion. Experiments demonstrate that ReCAP substantially improves subgoal alignment and success rates on various long-horizon reasoning benchmarks, achieving a 32\% gain on synchronous Robotouille and a 29\% improvement on asynchronous Robotouille under the strict pass@1 protocol.

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cyber-capabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a given vulnerability), and Patch (patching a given vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards from \\$10 to \\$30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a given vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet Thinking, Qwen3 235B A22B, Llama 4 Maverick, and DeepSeek-R1. Given up to three attempts, the top-performing agents are OpenAI Codex CLI: o3-high (12.5% on Detect, mapping to \\$3,720; 90% on Patch, mapping to \\$14,152), Custom Agent with Claude 3.7 Sonnet Thinking (67.5% on Exploit), and OpenAI Codex CLI: o4-mini (90% on Patch, mapping to \\$14,422). OpenAI Codex CLI: o3-high, OpenAI Codex CLI: o4-mini, and Claude Code are more capable at defense, achieving higher Patch scores of 90%, 90%, and 87.5%, compared to Exploit scores of 47.5%, 32.5%, and 57.5% respectively; while the custom agents are relatively balanced between offense and defense, achieving Exploit scores of 17.5-67.5% and Patch scores of 25-60%.

The rapid evolution of the Web as a key platform for information dissemination has led to the growing integration of large language models (LLMs) in Web-based applications. However, the swift changes in web content present challenges in maintaining these models' relevance and accuracy. The task of Knowledge Editing (KE) is aimed at efficiently and precisely adjusting the behavior of large language models (LLMs) to update specific knowledge while minimizing any adverse effects on other knowledge. Current research predominantly concentrates on editing white-box LLMs, neglecting a significant scenario: editing black-box LLMs, where access is limited to interfaces and only textual output is provided. In this paper, we initially officially introduce KE on black-box LLMs, followed by presenting a thorough evaluation framework. This framework operates without requiring logits and considers pre- and post-edit consistency, addressing the limitations of current evaluations that are inadequate for black-box LLMs editing and lack comprehensiveness. To address privacy leaks of editing data and style over-editing in existing approaches, we propose a new postEdit framework. postEdit incorporates a retrieval mechanism for editing knowledge and a purpose-trained editing plugin called post-editor, ensuring privacy through downstream processing and maintaining textual style consistency via fine-grained editing. Experiments and analysis conducted on two benchmarks show that postEdit surpasses all baselines and exhibits robust generalization, notably enhancing style retention by an average of +20.82%. Our code is available on github https://github.com/songxiaoshuai/postEdit.

Abstract In our study, we propose a self-supervised neural topic model (NTM) that combines the power of NTMs and regularized self-supervised learning methods to improve performance. NTMs use neural networks to learn latent topics hidden behind the words in documents, enabling greater flexibility and the ability to estimate more coherent topics compared to traditional topic models. On the other hand, some self-supervised learning methods use a joint embedding architecture with two identical networks that produce similar representations for two augmented versions of the same input. Regularizations are applied to these representations to prevent collapse, which would otherwise result in the networks outputting constant or redundant representations for all inputs. Our model enhances topic quality by explicitly regularizing latent topic representations of anchor and positive samples. We also introduced an adversarial data augmentation method to replace the heuristic sampling method. We further developed several variation models including those on the basis of an NTM that incorporates contrastive learning with both positive and negative samples. Experimental results on three datasets showed that our models outperformed baselines and state-of-the-art models both quantitatively and qualitatively.

Yejie Wang, Keqing He, Guanting Dong, Pei Wang, Weihao Zeng, Muxi Diao, Weiran Xu, Jingang Wang, Mengdi Zhang, Xunliang Cai. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Out-of-domain (OOD) intent detection aims to examine whether the user’s query falls outside the predefined domain of the system, which is crucial for the proper functioning of task-oriented dialogue (TOD) systems. Previous methods address it by fine-tuning discriminative models. Recently, some studies have been exploring the application of large language models (LLMs) represented by ChatGPT to various downstream tasks, but it is still unclear for their ability on OOD detection task.This paper conducts a comprehensive evaluation of LLMs under various experimental settings, and then outline the strengths and weaknesses of LLMs. We find that LLMs exhibit strong zero-shot and few-shot capabilities, but is still at a disadvantage compared to models fine-tuned with full resource. More deeply, through a series of additional analysis experiments, we discuss and summarize the challenges faced by LLMs and provide guidance for future work including injecting domain knowledge, strengthening knowledge transfer from IND(In-domain) to OOD, and understanding long instructions.

Pre-trained language models have been successful in many scenarios. However, their usefulness in task-oriented dialogues is limited due to the intrinsic linguistic differences between general text and task-oriented dialogues. Current task-oriented dialogue pre-training methods rely on a contrastive framework, which faces challenges such as selecting true positives and hard negatives, as well as lacking diversity. In this paper, we propose a novel dialogue pre-training model called BootTOD. It learns task-oriented dialogue representations via a self-bootstrapping framework. Unlike contrastive counterparts, BootTOD aligns context and context+response representations and dismisses the requirements of contrastive pairs. BootTOD also uses multiple appropriate response targets to model the intrinsic one-to-many diversity of human conversations. Experimental results show that BootTOD outperforms strong TOD baselines on diverse downstream dialogue tasks.

Yejie Wang, Keqing He, Dayuan Fu, Zhuoma GongQue, Heyang Xu, Yanxu Chen, Zhexu Wang, Yujia Fu, Guanting Dong, Muxi Diao, Jingang Wang, Mengdi Zhang, Xunliang Cai, Weiran Xu. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

In a realistic dialogue system, the input information from users is often subject to various types of input perturbations, which affects the slot-filling task. Although rule-based data augmentation methods have achieved satisfactory results, they fail to exhibit the desired generalization when faced with unknown noise disturbances. In this study, we address the challenges posed by input perturbations in slot filling by proposing Noise-BERT, a unified Perturbation-Robust Framework with Noise Alignment Pre-training. Our framework incorporates two Noise Alignment Pre-training tasks: Slot Masked Prediction and Sentence Noisiness Discrimination, aiming to guide the pre-trained language model in capturing accurate slot information and noise distribution. During fine-tuning, we employ a contrastive learning loss to enhance the semantic representation of entities and labels. Additionally, we introduce an adversarial attack training strategy to improve the model’s robustness. Experimental results demonstrate the superiority of our proposed approach over state-of-the-art models, and further analysis confirms its effectiveness and generalization ability.

Yutao Mou, Xiaoshuai Song, Keqing He, Chen Zeng, Pei Wang, Jingang Wang, Yunsen Xian, Weiran Xu. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Weihao Zeng, Keqing He, Yejie Wang, Chen Zeng, Jingang Wang, Yunsen Xian, Weiran Xu. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Existing controllable dialogue generation work focuses on the single-attribute control and lacks generalization capability to out-of-distribution multiple attribute combinations.In this paper, we explore the compositional generalization for multi-attribute controllable dialogue generation where a model can learn from seen attribute values and generalize to unseen combinations.We propose a prompt-based disentangled controllable dialogue generation model, DCG.It learns attribute concept composition by generating attribute-oriented prompt vectors and uses a disentanglement loss to disentangle different attributes for better generalization.Besides, we design a unified reference-free evaluation framework for multiple attributes with different levels of granularities.Experiment results on two benchmarks prove the effectiveness of our method and the evaluation metric.* The first two authors contribute equally.Weiran Xu is the corresponding author.

Knowledge Base Question Answering (KBQA) aims to answer natural language questions with factual information such as entities and relations in KBs. However, traditional Pre-trained Language Models (PLMs) are directly pre-trained on large-scale natural language corpus, which poses challenges for them in understanding and representing complex subgraphs in structured KBs. To bridge the gap between texts and structured KBs, we propose a Structured Knowledge-aware Pre-training method (SKP). In the pre-training stage, we introduce two novel structured knowledge-aware tasks, guiding the model to effectively learn the implicit relationship and better representations of complex subgraphs. In downstream KBQA task, we further design an efficient linearization strategy and an interval attention mechanism, which assist the model to better encode complex subgraphs and shield the interference of irrelevant subgraphs during reasoning respectively. Detailed experiments and analyses on WebQSP verify the effectiveness of SKP, especially the significant improvement in subgraph retrieval (+4.08% H@10).

The objective of few-shot named entity recognition is to identify named entities with limited labeled instances. Previous works have primarily focused on optimizing the traditional token-wise classification framework, while neglecting the exploration of information based on NER data characteristics. To address this issue, we propose a Multi-Task Semantic Decomposition Framework via Joint Task-specific Pre-training (MSDP) for few-shot NER. Drawing inspiration from demonstration-based and contrastive learning, we introduce two novel pre-training tasks: Demonstration-based Masked Language Modeling (MLM) and Class Contrastive Discrimination. These tasks effectively incorporate entity boundary information and enhance entity representation in Pre-trained Language Models (PLMs). In the downstream main task, we introduce a multi-task joint optimization framework with the semantic decomposing method, which facilitates the model to integrate two different semantic information for entity classification. Experimental results of two few-shot NER benchmarks demonstrate that MSDP consistently outperforms strong baselines by a large margin. Extensive analyses validate the effectiveness and generalization of MSDP.

Xiaoshuai Song, Keqing He, Pei Wang, Guanting Dong, Yutao Mou, Jingang Wang, Yunsen Xian, Xunliang Cai, Weiran Xu. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Few-shot named entity recognition (NER) aims at identifying named entities based on only few labeled instances. Most existing prototype-based sequence labeling models tend to memorize entity mentions which would be easily confused by close prototypes. In this paper, we proposed a Prototypical Semantic Decoupling method via joint Contrastive learning (PSDC) for few-shot NER. Specifically, we decouple class-specific prototypes and contextual semantic prototypes by two masking strategies to lead the model to focus on two different semantic information for inference. Besides, we further introduce joint contrastive learning objectives to better integrate two kinds of decoupling information and prevent semantic collapse. Experimental results on two few-shot NER benchmarks demonstrate that PSDC consistently outperforms the previous SOTA methods in terms of overall performance. Extensive analysis further validates the effectiveness and generalization of PSDC.

In real dialogue scenarios, the existing slot filling model, which tends to memorize entity patterns, has a significantly reduced generalization facing Out-of-Vocabulary (OOV) problems. To address this issue, we propose an OOV robust slot filling model based on multi-level data augmentations to solve the OOV problem from both word and slot perspectives. We present a unified contrastive learning framework, which pull representations of the origin sample and augmentation samples together, to make the model resistant to OOV problems. We evaluate the performance of the model from some specific slots and carefully design test data with OOV word perturbation to further demonstrate the effectiveness of OOV words. Experiments on two datasets show that our approach outperforms the previous sota methods in terms of both OOV slots and words.

Most existing slot filling models tend to memorize inherent patterns of entities and corresponding contexts from training data. However, these models can lead to system failure or undesirable outputs when being exposed to spoken language perturbation or variation in practice. We propose a perturbed semantic structure awareness transferring method for training perturbation-robust slot filling models. Specifically, we introduce two MLM-based training strategies to respectively learn contextual semantic structure and word distribution from unsupervised language perturbation corpus. Then, we transfer semantic knowledge learned from upstream training procedure into the original samples and filter generated data by consistency processing. These procedures aims to enhance the robustness of slot filling models. Experimental results show that our method consistently outperforms the previous basic methods and gains strong generalization while preventing the model from memorizing inherent patterns of entities and contexts.

Traditional intent classification models are based on a pre-defined intent set and only recognize limited in-domain (IND) intent classes. But users may input out-of-domain (OOD) queries in a practical dialogue system. Such OOD queries can provide directions for future improvement. In this paper, we define a new task, Generalized Intent Discovery (GID), which aims to extend an IND intent classifier to an open-world intent set including IND and OOD intents. We hope to simultaneously classify a set of labeled IND intent classes while discovering and recognizing new unlabeled OOD types incrementally. We construct three public datasets for different application scenarios and propose two kinds of frameworks, pipeline-based and end-to-end for future work. Further, we conduct exhaustive experiments and qualitative analysis to comprehend key challenges and provide new guidance for future GID research.

Out-of-Domain (OOD) detection is a key component in a task-oriented dialog system, which aims to identify whether a query falls outside the predefined supported intent set. Previous softmax-based detection algorithms are proved to be overconfident for OOD samples. In this paper, we analyze overconfident OOD comes from distribution uncertainty due to the mismatch between the training and test distributions, which makes the model can’t confidently make predictions thus probably causes abnormal softmax scores. We propose a Bayesian OOD detection framework to calibrate distribution uncertainty using Monte-Carlo Dropout. Our method is flexible and easily pluggable to existing softmax-based baselines and gains 33.33% OOD F1 improvements with increasing only 0.41% inference time compared to MSP. Further analyses show the effectiveness of Bayesian learning for OOD detection.

Collecting dialogue data with domain-slotvalue labels for dialogue state tracking (DST) could be a costly process.In this paper, we propose a novel framework based on domain-slot related description to tackle the challenge of few-shot cross-domain DST.Specifically, we design an extraction module to extract domainslot related verbs and nouns in the dialogue.Then, we integrates them into the description, which aims to prompt the model to identify the slot information.Furthermore, we introduce a random sampling strategy to improve the domain generalization ability of the model.We utilize a pre-trained model to encode contexts and description and generates answers with an auto-regressive manner.Experimental results show that our approaches substantially outperform the existing few-shot DST methods on MultiWOZ and gain strong improvements on the slot accuracy comparing to existing slot description methods.

Discovering out-of-domain (OOD) intent is important for developing new skills in taskoriented dialogue systems.The key challenges lie in how to transfer prior in-domain (IND) knowledge to OOD clustering, as well as jointly learn OOD representations and cluster assignments.Previous methods suffer from indomain overfitting problem, and there is a natural gap between representation learning and clustering objectives.In this paper, we propose a unified K-nearest neighbor contrastive learning framework to discover OOD intents.Specifically, for IND pre-training stage, we propose a KCL objective to learn inter-class discriminative features, while maintaining intraclass diversity, which alleviates the in-domain overfitting problem.For OOD clustering stage, we propose a KCC method to form compact clusters by mining true hard negative samples, which bridges the gap between clustering and representation learning.Extensive experiments on three benchmark datasets show that our method achieves substantial improvements over the state-of-the-art methods. 1

Detecting out-of-domain (OOD) intents from user queries is essential for avoiding wrong operations in task-oriented dialogue systems.The key challenge is how to distinguish indomain (IND) and OOD intents.Previous methods ignore the alignment between representation learning and scoring function, limiting the OOD detection performance.In this paper, we propose a unified neighborhood learning framework (UniNL) to detect OOD intents.Specifically, we design a K-nearest neighbor contrastive learning (KNCL) objective for representation learning and introduce a KNNbased scoring function for OOD detection.We aim to align representation learning with scoring function.Experiments and analysis on two benchmark datasets show the effectiveness of our method.1

Multi-domain text classification can automatically classify texts in various scenarios. Due to the diversity of human languages, texts with the same label in different domains may differ greatly, which brings challenges to the multi-domain text classification. Current advanced methods use the private-shared paradigm, capturing domain-shared features by a shared encoder, and training a private encoder for each domain to extract domain-specific features. However, in realistic scenarios, these methods suffer from inefficiency as new domains are constantly emerging. In this paper, we propose a robust contrastive alignment method to align text classification features of various domains in the same feature space by supervised contrastive learning. By this means, we only need two universal feature extractors to achieve multi-domain text classification. Extensive experimental results show that our method performs on par with or sometimes better than the state-of-the-art method, which uses the complex multi-classifier in a private-shared framework.

Yanan Wu, Keqing He, Yuanmeng Yan, QiXiang Gao, Zhiyuan Zeng, Fujia Zheng, Lulu Zhao, Huixing Jiang, Wei Wu, Weiran Xu. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Lulu Zhao, Fujia Zheng, Weihao Zeng, Keqing He, Weiran Xu, Huixing Jiang, Wei Wu, Yanan Wu. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Traditional dialogue summarization models rely on a large-scale manually-labeled corpus, lacking generalization ability to new domains, and domain adaptation from a labeled source domain to an unlabeled target domain is important in practical summarization scenarios. However, existing domain adaptation works in dialogue summarization generally require large-scale pre-training using extensive external data. To explore the lightweight fine-tuning methods, in this paper, we propose an efficient Adversarial Disentangled Prompt Learning (ADPL) model for domain adaptation in dialogue summarization. We introduce three kinds of prompts including domain-invariant prompt (DIP), domain-specific prompt (DSP), and task-oriented prompt (TOP). DIP aims to disentangle and transfer the shared knowledge from the source domain and target domain in an adversarial way, which improves the accuracy of prediction about domain-invariant information and enhances the ability for generalization to new domains. DSP is designed to guide our model to focus on domain-specific knowledge using domain-related features. TOP is to capture task-oriented knowledge to generate high-quality summaries. Instead of fine-tuning the whole pre-trained language model (PLM), we only update the prompt networks but keep PLM fixed. Experimental results on the zero-shot setting show that the novel design of prompts can yield more coherent, faithful, and relevant summaries than baselines using the prefix-tuning, and perform at par with fine-tuning while being more efficient. Overall, our work introduces a prompt-based perspective to the zero-shot learning for dialogue summarization task and provides valuable findings and insights for future research.

Yanan Wu, Zhiyuan Zeng, Keqing He, Hong Xu, Yuanmeng Yan, Huixing Jiang, Weiran Xu. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Yuanmeng Yan, Rumei Li, Sirui Wang, Fuzheng Zhang, Wei Wu, Weiran Xu. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Zero-shot cross-domain slot filling alleviates the data dependence in the case of data scarcity in the target domain, which has aroused extensive research.However, as most of the existing methods do not achieve effective knowledge transfer to the target domain, they just fit the distribution of the seen slot and show poor performance on unseen slot in the target domain.To solve this, we propose a novel approach based on prototypical contrastive learning with a dynamic label confusion strategy for zero-shot slot filling.The prototypical contrastive learning aims to reconstruct the semantic constraints of labels, and we introduce the label confusion strategy to establish the label dependence between the source domains and the target domain on-the-fly.Experimental results show that our model achieves significant improvement on the unseen slots, while also set new state-of-the-arts on slot filling task. 1

The key challenge of question answering over knowledge bases (KBQA) is the inconsistency between the natural language questions and the reasoning paths in the knowledge base (KB).Recent graph-based KBQA methods are good at grasping the topological structure of the graph but often ignore the textual information carried by the nodes and edges.Meanwhile, pre-trained language models learn massive open-world knowledge from the large corpus, but it is in the natural language form and not structured.To bridge the gap between the natural language and the structured KB, we propose three relation learning tasks for BERTbased KBQA, including relation extraction, relation matching, and relation reasoning.By relation-augmented training, the model learns to align the natural language expressions to the relations in the KB as well as reason over the missing connections in the KB.Experiments on WebQSP show that our method consistently outperforms other baselines, especially when the KB is incomplete.

Neural abstractive summarization systems have gained significant progress in recent years.However, abstractive summarization often produce inconsisitent statements or false facts.How to automatically generate highly abstract yet factually correct summaries?In this paper, we proposed an efficient weaksupervised adversarial data augmentation approach to form the factual consistency dataset.Based on the artificial dataset, we train an evaluation model that can not only make accurate and robust factual consistency discrimination but is also capable of making interpretable factual errors tracing by backpropagated gradient distribution on token embeddings.Experiments and analysis conduct on public annotated summarization and factual consistency datasets demonstrate our approach effective and reasonable.Our codes can be found at https://github.com/ parZival27/GrAdualCC

Traditional document summarization models cannot handle dialogue summarization tasks perfectly. In situations with multiple speakers and complex personal pronouns referential relationships in the conversation. The predicted summaries of these models are always full of personal pronoun confusion. In this paper, we propose a hierarchical transformer-based model for dialogue summarization. It encodes dialogues from words to utterances and distinguishes the relationships between speakers and their corresponding personal pronouns clearly. In such a from-coarse-to-fine procedure, our model can generate summaries more accurately and relieve the confusion of personal pronouns. Experiments are based on a dialogue summarization dataset SAMsum, and the results show that the proposed model achieved a comparable result against other strong baselines. Empirical experiments have shown that our method can relieve the confusion of personal pronouns in predicted summaries.

Detecting out-of-domain (OOD) intents is critical in a task-oriented dialog system. Existing methods rely heavily on extensive manually labeled OOD samples and lack robustness. In this paper, we propose an efficient adversarial attack mechanism to augment hard OOD samples and design a novel generative distance-based classifier to detect OOD samples instead of a traditional threshold-based discriminator classifier. Experiments on two public benchmark datasets show that our method can consistently outperform the baselines with a statistically significant margin.

The performance of many intent detection approaches will degrade when they meet open-set data because there is out-of-domain (OOD) noise. Some works utilize clean but expensive labeled data to supervise models more robust to the varied environment. However, a large number of labeled samples are scarce and their models no longer change after finishing training. To address this problem, we propose an iterative learning framework that can dynamically improve the model's ability of OOD intent detection and meanwhile continually obtain valuable new data to learn deep discriminative features. Concretely, the model can generate pseudo labels for unlabeled examples by self-training and use the local outlier factor (LOF) algorithm to detect unknown intents. Furthermore, we add mask operation on supervised contrastive loss (SCL) and use the masked-SCL to absorb new data effectively. Experiments on CLINC and SNIPS demonstrate that our proposed method can robustly realize intent detection in the presence of a high proportion of open-set.

Liwen Wang, Yuanmeng Yan, Keqing He, Yanan Wu, Weiran Xu. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Zhiyuan Zeng, Keqing He, Yuanmeng Yan, Hong Xu, Weiran Xu. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Large-scale knowledge graphs (KGs) are shown to become more important in current information systems. To expand the coverage of KGs, previous studies on knowledge graph completion need to collect adequate training instances for newly-added relations. In this paper, we consider a novel formulation, zero-shot learning, to free this cumbersome curation. For newly-added relations, we attempt to learn their semantic features from their text descriptions and hence recognize the facts of unseen relations with no examples being seen. For this purpose, we leverage Generative Adversarial Networks (GANs) to establish the connection between text and knowledge graph domain: The generator learns to generate the reasonable relation embeddings merely with noisy text descriptions. Under this setting, zero-shot learning is naturally converted to a traditional supervised classification task. Empirically, our method is model-agnostic that could be potentially applied to any version of KG embeddings, and consistently yields performance improvements on NELL and Wiki dataset.

Neural-based context-aware models for slot tagging have achieved state-of-the-art performance.However, the presence of OOV(outof-vocab) words significantly degrades the performance of neural-based models, especially in a few-shot scenario.In this paper, we propose a novel knowledge-enhanced slot tagging model to integrate contextual representation of input text and the large-scale lexical background knowledge.Besides, we use multilevel graph attention to explicitly model lexical relations.The experiments show that our proposed knowledge integration mechanism achieves consistent improvements across settings with different sizes of training data on two public benchmark datasets.

Contextualized neural language models have gained much attention in Information Retrieval (IR) with its ability to achieve better text understanding by capturing contextual structure. However, to achieve better document understanding, it is necessary to involve global structure of a document. In this paper, we take the advantage of Graph Convolutional Networks (GCN) to model global word-relation structure of a document to improve context-aware document ranking. We propose to build a graph for a document to model the global structure. The nodes and edges of the graph are constructed from contextual embeddings. Then we apply graph convolution on the graph to learning a new representation, and this representation covers both contextual and global structure information. The experimental results show that our method outperforms the state-of-the-art contextual language models, which demonstrate that incorporating global structure is useful for improving document ranking and GCN is an effective way to achieve it.

Zero-shot slot filling has widely arisen to cope with data scarcity in target domains.However, previous approaches often ignore constraints between slot value representation and related slot description representation in the latent space and lack enough model robustness.In this paper, we propose a Contrastive Zero-Shot Learning with Adversarial Attack (CZSL-Adv) method for the cross-domain slot filling.The contrastive loss aims to map slot value contextual representations to the corresponding slot description representations.And we introduce an adversarial attack training strategy to improve model robustness.Experimental results show that our model significantly outperforms state-of-the-art baselines under both zero-shot and few-shot settings.

Detecting out-of-domain (OOD) input intents is critical in the task-oriented dialog system. Different from most existing methods that rely heavily on manually labeled OOD samples, we focus on the unsupervised OOD detection scenario where there are no labeled OOD samples except for labeled in-domain data. In this paper, we propose a simple but strong generative distance-based classifier to detect OOD samples. We estimate the class-conditional distribution on feature spaces of DNNs via Gaussian discriminant analysis (GDA) to avoid over-confidence problems. And we use two distance functions, Euclidean and Mahalanobis distances, to measure the confidence score of whether a test sample belongs to OOD. Experiments on four benchmark datasets show that our method can consistently outperform the baselines.

Recently, people have been beginning paying more attention to the abstractive dialogue summarization task. Since the information flows are exchanged between at least two interlocutors and key elements about a certain event are often spanned across multiple utterances, it is necessary for researchers to explore the inherent relations and structures of dialogue contents. However, the existing approaches often process the dialogue with sequence-based models, which are hard to capture long-distance inter-sentence relations. In this paper, we propose a Topic-word Guided Dialogue Graph Attention (TGDGA) network to model the dialogue as an interaction graph according to the topic word information. A masked graph self-attention mechanism is used to integrate cross-sentence information flows and focus more on the related utterances, which makes it better to understand the dialogue. Moreover, the topic word features are introduced to assist the decoding process. We evaluate our model on the SAMSum Corpus and Automobile Master Corpus. The experimental results show that our method outperforms most of the baselines.

Open-vocabulary slots, such as file name, album name, or schedule title, significantly degrade the performance of neural-based slot filling models since these slots can take on values from a virtually unlimited set and have no semantic restriction nor a length limit.In this paper, we propose a robust adversarial model-agnostic slot filling method that explicitly decouples local semantics inherent in open-vocabulary slot words from the global context.We aim to depart entangled contextual semantics and focus more on the holistic context at the level of the whole sentence.Experiments on two public datasets show that our method consistently outperforms other methods with a statistically significant margin on all the open-vocabulary slots without deteriorating the performance of normal slots. *The first two authors contribute equally.Weiran Xu is the corresponding author.

In task-oriented dialogue systems, dialogue state tracking (DST) is an essential part which aims to estimate user goal at every turn 1 of the dialogue. At each turn, DST aims to estimate user goals by current user utterance and last system action. However, most current approaches encode these relevant sequences by recurrent networks, which is a challenge for these models to capture long-range dependencies. Besides predicting the current dialogue state is relevant to historical context and sometimes refers to the past utterances. In this paper, we propose the Gated Attentive Convolutional network Dialogue State Tracker (GAC) which overcomes these challenges by utilizing the gated attentive convolutional encoder and introducing historical information. We apply a gated attentive convolutional network encoder to learn the sequence representation and introduce historical dialogue utterances as evidence to track the user goal. Experiments show that our method improves joint goal accuracy on WoZ 2.0[1] and MultiWoZ 2.0[2] datasets respectively outperforms the previous state-of-the-art GLAD [3] and Trade [4] model1.

Slot tagging is a key component in a task-oriented dialogue system. Conversational agents need to understand human input by training on large amounts of annotated data. However, most human languages are low-resource and lack annotated training data for slot tagging task. Therefore, we aim to leverage cross-lingual transfer learning from high-resource languages to low-resource ones. In this paper, we propose an adversarial cross-lingual transfer model with multi-level language shared and specific knowledge to improve the slot tagging task of low-resource languages. Our method explicitly separates the model into the language-shared part and language-specific part to transfer language-independent knowledge. To refine shared knowledge in the latent space, we add a language discriminator and employ adversarial training to reinforce feature separation. Besides, we adopt a novel multi-level feature transfer in an incremental and progressive way to acquire multi-granularity shared knowledge. To mitigate the discrepancies between the feature distributions of language specific and shared knowledge, we propose the neural adapters to fuse features from different sources. Experiments show that our proposed model consistently outperforms monolingual baseline with a statistically significant margin up to 2.09%, even higher improvement of 12.21% in the zero-shot setting. Further analysis demonstrates that our method could effectively alleviate data scarcity of low-resource languages.

Sequence labeling is a fundamental task of natural language understanding. Recent neural models for sequence labeling task achieve significant success with the availability of sufficient training data. However, in practical scenarios, entity types to be annotated even in the same domain are continuously evolving. To transfer knowledge from the source model pre-trained on previously annotated data, we propose an approach which learns label-relational output structure to explicitly capturing label correlations in the latent space. Additionally, we construct the target-to-source interaction between the source model MSand the target model MTand apply a gate mechanism to control how much information in MSand MTshould be passed down. Experiments show that our method consistently outperforms the state-of-the-art methods with a statistically significant margin and effectively facilitates to recognize rare new entities in the target data especially.

Contextualized neural language models have gained much attention in Information Retrieval (IR) with its ability to achieve better word understanding by capturing contextual structure on sentence level. However, to understand a document better, it is necessary to involve contextual structure from document level. Moreover, some words contributes more information to delivering the meaning of a document. Motivated by this, in this paper, we take the advantages of Graph Convolutional Networks (GCN) and Graph Attention Networks (GAN) to model global word-relation structure of a document with attention mechanism to improve context-aware document ranking. We propose to build a graph for a document to model the global contextual structure. The nodes and edges of the graph are constructed from contextual embeddings. We first apply graph convolution on the graph and then use attention networks to explore the influence of more informative words to obtain a new representation. This representation covers both local contextual and global structure information. The experimental results show that our method outperforms the state-of-the-art contextual language models, which demonstrate that incorporating contextual structure is useful for improving document ranking.

We propose a salient-context based semantic matching method to improve relevance ranking in information retrieval. We first propose a new notion of salient context and then define how to measure it. Then we show how the most salient context can be located with a sliding window technique. Finally, we use the semantic similarity between a query term and the most salient context terms in a corpus of documents to rank those documents. Experiments on various TREC collections show the effectiveness of our model compared to the state-of-the-art methods.

Distant supervision can effectively label data for relation extraction, but suffers from the noise labeling problem.Recent works mainly perform soft bag-level noise reduction strategies to find the relatively better samples in a sentence bag, which is suboptimal compared with making a hard decision of false positive samples in sentence level.In this paper, we introduce an adversarial learning framework, which we named DSGAN, to learn a sentencelevel true-positive generator.Inspired by Generative Adversarial Networks, we regard the positive samples generated by the generator as the negative samples to train the discriminator.The optimal generator is obtained until the discrimination ability of the discriminator has the greatest decline.We adopt the generator to filter distant supervision training dataset and redistribute the false positive instances into the negative set, in which way to provide a cleaned dataset for relation classification.The experimental results show that the proposed strategy significantly improves the performance of distant supervision relation extraction comparing to state-of-the-art systems.

Distant supervision has become the standard method for relation extraction.However, even though it is an efficient method, it does not come at no cost-The resulted distantly-supervised training samples are often very noisy.To combat the noise, most of the recent state-of-theart approaches focus on selecting onebest sentence or calculating soft attention weights over the set of the sentences of one specific entity pair.However, these methods are suboptimal, and the false positive problem is still a key stumbling bottleneck for the performance.We argue that those incorrectly-labeled candidate sentences must be treated with a hard decision, rather than being dealt with soft attention weights.To do this, our paper describes a radical solution-We explore a deep reinforcement learning strategy to generate the false-positive indicator, where we automatically recognize false positives for each relation type without any supervised information.Unlike the removal operation in the previous studies, we redistribute them into the negative examples.The experimental results show that the proposed strategy significantly improves the performance of distant supervision comparing to state-of-the-art systems.

Entity pair provide essential information for identifying relation type. Aiming at this characteristic, Position Feature is widely used in current relation classification systems to highlight the words close to them. However, semantic knowledge involved in entity pair has not been fully utilized. To overcome this issue, we propose an Entity-pair-based Attention Mechanism, which is specially designed for relation classification. Recently, attention mechanism significantly promotes the development of deep learning in NLP. Inspired by this, for specific instance(entity pair, sentence), the corresponding entity pair information is incorporated as prior knowledge to adaptively compute attention weights for generating sentence representation. Experimental results on SemEval-2010 Task 8 dataset show that our method outperforms most of the state-of-the-art models, without external linguistic features.

Annotation projection is a practical method to deal with the low resource problem in incident languages (IL) processing. Previous methods on annotation projection mainly relied on word alignment results without any training process, which led to noise propagation caused by word alignment errors. In this paper, we focus on the named entity recognition (NER) task and propose a weakly-supervised framework to project entity annotations from English to IL through bitexts. Instead of directly relying on word alignment results, this framework combines advantages of rule-based methods and deep learning methods by implementing two steps: First, generates a high-confidence entity annotation set on IL side with strict searching methods; Second, uses this high-confidence set to weakly supervise the model training. The model is finally used to accomplish the projecting process. Experimental results on two low-resource ILs show that the proposed method can generate better annotations projected from English-IL parallel corpora. The performance of IL name tagger can also be improved significantly by training on the newly projected IL annotation set.

We propose some better word embedding models based on vLBL model and ivLBL model by sharing representations between context and target words and using document representations. Our proposed models are much simpler which have almost half less parameters than the state-of-the-art methods. We achieve better results on word analogy task than the best ones reported before using significantly less training data and computing time.

Extracting product aspects and their associated sentiments is one of the key tasks in sentiment analysis. Estimating the confidences of extracted aspects is important to ensure the performance. To tackle the issue, this paper proposes a two-step estimation method. Collocations of product features and opinion words are initially extracted through pattern bootstrapping. A criterion synthesizing two measurements, Popularity and Reliability, is novelly exploited to assess both patterns and features. Then the features are further clustered into aspects based on path similarities in the Word Net. Each cluster is assigned a weight based on its Compactness and Texture, and the light ones are filtered out. In addition, this paper also captures global aspect reputations by aggregating sentiment strengths through opinion collocations. Experimental results on a benchmark data set with 5 products demonstrate the effectiveness and reliability of our proposed method.

In this paper, we propose a time-aware approach to promoting textual retrieval performance for Electronic Health Record (EHR) search. The proposed approach focuses on identifying patients cohorts from the perspective of EHR temporal correlation. First, an EHR temporal profile is created according to EHR distribution on time interval for each patient. Second, the temporal similarity is computed and used as a feature for discovering temporal cohorts. In each cohort, the highest-ranked profile in textual retrieval is considered as the centroid, and a temporal relevance score is computed by multiplying temporal similarity with the textual relevance of the centroid. Finally, the temporal relevance is combined linearly with the textual relevance for re-ranking. Extensive experiments are conducted to demonstrate the effectiveness of the proposed approach in promoting retrieval performance for EHR search.

To alleviate the workload of labeling before estimating certain color distributions, integrative labeling is introduced, which merely needs to figure out whether a picture contains positive-class regions or not and then all pixels of the picture are treated as positive or negative class training samples. Integrative labeling, however, results in heavy mixture of training samples. Thus traditional generative density estimation methods can't be used directly in that they perform poorly with heavily polluted training samples. In this paper, by utilizing the prior knowledge of high separability between positive and negative class color distributions, a discriminative learning based GMM(DiscGMM) is proposed for integrative labeling. Besides generating the polluted positive-class samples with comparatively high probability, optimal parameters found by DiscGMM also enjoy a comparatively low probability of generating negative-class samples. The parameter learning problem is solved by a modified Expectation Maximization (EM) algorithm. In an integrative labeling experiment of skin detection, DiscGMM is testified to enjoy much better performance than generative density estimation methods and shows qualified results.

This paper proposes a novel Product Comparison approach. The comparative relations between products are first mined from both user reviews on multiple review websites and community-based question answering pairs containing product comparison information. A unified graph model is then developed to integrate the resultant comparative relations for product comparison. Experiments on popular electronic products show that the proposed approach outperforms the state-of-the-art methods.

This paper focuses on the task of text sentiment analysis in hybrid online articles and web pages. Traditional approaches of text sentiment analysis typically work at a particular level, such as phrase, sentence or document level, which might not be suitable for the documents with too few or too many words. Considering every level analysis has its own advantages, we expect that a combination model may achieve better performance. In this paper, a novel combined model based on phrase and sentence level's analyses and a discussion on the complementation of different levels' analyses are presented. For the phrase-level sentiment analysis, a newly defined Left-Middle-Right template and the Conditional Random Fields are used to extract the sentiment words. The Maximum Entropy model is used in the sentence-level sentiment analysis. The experiment results verify that the combination model with specific combination of features is better than single level model.

This paper introduces a locality discriminating indexing (LDI) algorithm for document classification. Based on the hypothesis that samples from different classes reside in class-specific manifold structures, LDI seeks for a projection which best preserves the within-class local structures while suppresses the between-class overlap. Comparative experiments show that the proposed method isable to derives compact discriminating document representations for classification.