YiJie Huang, Xiaocui Yang, Shi Feng, Wen Zhang, Kaisong Song, Yifei Zhang, Daling Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Ensuring fairness in social survey simulation is critical, as biased outputs can misrepresent underrepresented groups.This issue is growing as large language models (LLMs) are increasingly used for this task.However, standard finetuning based on Empirical Risk Minimization (ERM) often under-optimizes minority groups, causing substantial subgroup disparities.Distributionally robust Optimization (DRO) methods reduce worst-case errors, but their strict worst-case selection can lead to noisy and unstable optimization under demographic sparsity.These issues create intertwined challenges for fairness, convergence and stability.We propose SAFO, a dynamic utility-fairness optimization framework for LLM-based survey simulation that explicitly targets both fairness and training stability.SAFO combines (i) an Optimizer that preserves mean-loss utility, (ii) an Adversary that performs temperature-controlled, EMAsmoothed and loss-driven group reweighting, and (iii) a Nash-inspired Regulator that adaptively adjusts the utility-fairness trade-off by tracking weak-group gains and collateral utility damages.Experiments on three large-scale survey datasets from China, the U.S., and Europe show that SAFO consistently improves minority performance and social-welfare metrics.It reduces worst-group gaps by up to 12.7%, maintains overall accuracy with a mean change of less than 0.3% and lowers variance across random seeds.Our code is available at https://github.com/PiLab-ZJU/SAFO. * Majority (50-65) Acc.-Minority Acc.--Smaller is fairer Bias in Social Simulation Models Leads to Public Harm Majority Minority Social Survey Data (Imbalanced) Biased Model Fair Model Listens to Majority Accurately reflects all Minority

YongKang Liu, Jiayang Yu, Mingyang Wang, Yiqun Zhang, Ercong Nie, Shi Feng, Daling Wang, Kaisong Song, Hinrich Schuetze. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Weikang Yuan, Kaisong Song, Zhuoren Jiang, Junjie Cao, Yujie Zhang, Jun Lin, Kun Kuang, Ji Zhang, Xiaozhong Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Comparative Opinion Quintuple Extraction (COQE) aims to extract all comparative sentiment quintuples from product review text. Each quintuple comprises five elements: subject, object, aspect, opinion and preference. With the rise of Large Language Models (LLMs), existing work primarily focuses on enhancing the performance of COQE task through data augmentation, supervised fine-tuning and instruction tuning. Instead of the above pre-modeling and in-modeling design techniques, we focus on innovation in the post-processing. We introduce a model-unaware adaptive chain-of-feedback (COF) method from the perspective of inference feedback and extraction revision. This method comprises three core modules: dynamic example selection, self-critique and self-revision. By integrating LLMs, COF enables dynamic iterative self-optimization, making it applicable across different baselines. To validate the effectiveness of our approach, we utilize the outputs of two distinct baselines as inputs for COF: frozen parameters few-shot learning and the SOTA supervised fine-tuned model. We evaluate our approach on three benchmarks: Camera, Car and Ele. Experimental results show that, compared to the few-shot learning method, our approach achieves F1 score improvements of 3.51%, 2.65% and 5.28% for exact matching on the respective dataset. Even more impressively, our method further boosts performance, surpassing the current SOTA results, with additional gains of 0.76%, 6.54%, and 2.36% across the three datasets.

Large language models (LLMs) have shown significant potential to improve diagnostic performance for clinical professionals.Existing multi-agent paradigms rely mainly on prompt engineering, suffering from improper agent selection and insufficient knowledge integration.In this work, we propose a novel framework KACR (Knowledge-Aware Co-Reasoning) that integrates structured knowledge reasoning into multidisciplinary collaboration from two aspects: (1) a reinforcement learning-optimized agent that uses clinical knowledge graphs to guide dynamic discipline determination; (2) a multidisciplinary collaboration strategy that enables robust consensus through integration of domain-specific expertise and interdisciplinary persuasion mechanism.Extensive experiments conducted on both academic and real-world datasets demonstrate the effectiveness of our method.1 Main work done when working at Alibaba: https:// anonymous.4open.science/r/KACR_RL-2B64agents iteratively refine their positions through evidence-based persuasion.Persuasion strength is explicitly measured using the value estimates from the Critic trained in Step I, ensuring that consensus-building is aligned with the underlying clinical knowledge structure.This dual mechanism effectively balances specialized expertise with collective intelligence during differential diagnosis.Knowledge Graph.The clinical knowledge graph (CKG), denoted as G, comprises three core components: entity set V, structural relation set E, and relation type set R. The entities are classified into three distinct categories: symptom entities V s , disease entities V d , and discipline entities V c .Each relation is formally represented as a triplet (v i , r, v j ), where v i (head entity) and v j (tail entity) are interconnected through the relation type r ∈ R.

Large Language Models (LLMs) have demonstrated remarkable capabilities, yet their further evolution is often hampered by the scarcity of high-quality training data and the heavy reliance of traditional methods on expert-labeled data.This reliance sets a ceiling on LLM performance and is particularly challenging in low data resource scenarios where extensive supervision is unavailable.To address this issue, we propose a novel paradigm named LANCE (LANguage models as Continuous self-Evolving data engineers) that enables LLMs to train themselves by autonomously generating, cleaning, reviewing, and annotating data with preference information.Our approach demonstrates that LLMs can serve as continuous self-evolving data engineers, significantly reducing the time and cost of post-training data construction.Through iterative fine-tuning on Qwen2 series models, we validate the effectiveness of LANCE across various tasks, showing that it can maintain high-quality data generation and continuously improve model performance.Across multiple benchmark dimensions, LANCE results in an average score enhancement of 3.64 for Qwen2-7B and 1.75 for Qwen2-7B-Instruct.This autonomous data construction paradigm not only lessens reliance on human experts or external models but also ensures data aligns with human preferences, offering a scalable path for LLM self-improvement, especially in contexts with limited supervisory data.

While self-supervised graph pretraining techniques have shown promising results in various domains, their application still experiences challenges of limited topology learning, human knowledge dependency, and incompetent multi-level interactions. To address these issues, we propose a novel solution, Dual-level Graph self-supervised Pretraining with Motif discovery (DGPM), which introduces a unique dual-level pretraining structure that orchestrates node-level and subgraph-level pretext tasks. Unlike prior approaches, DGPM autonomously uncovers significant graph motifs through an edge pooling module, aligning learned motif similarities with graph kernel-based similarities. A cross-matching task enables sophisticated node-motif interactions and novel representation learning. Extensive experiments on 15 datasets validate DGPM's effectiveness and generalizability, outperforming state-of-the-art methods in unsupervised representation learning and transfer learning settings. The autonomously discovered motifs demonstrate the potential of DGPM to enhance robustness and interpretability.

Yiqun Zhang, Fanheng Kong, Peidong Wang, Shuang Sun, SWangLing SWangLing, Shi Feng, Daling Wang, Yifei Zhang, Kaisong Song. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Recently, we have witnessed the breakthroughs of meta-learning for few-shot learning scenario. Data augmentation is essential for meta-learning, particularly in situations where data is extremely scarce. However, existing text data augmentation methods can not ensure the diversity and quality of the generated data, which leads to sub-optimal performance. Inspired by the recent success of large language models (LLMs) which demonstrate improved language comprehension abilities, we propose a Meta-learning framework with Progressive Data Augmentation (PDAMeta) for few-shot text classification, which contains a two-stage data augmentation strategy. First, the prompt-based data augmentation enriches the diversity of the training instances from a global perspective. Second, the attention-based data augmentation further improves the data quality from a local perspective. Last, we propose a dual-stream contrastive meta-learning strategy to learn discriminative text representations from both original and augmented instances. Extensive experiments conducted on four public few-shot text classification datasets show that PDAMeta significantly outperforms several state-of-the-art models and shows better robustness.

Scientific Information Extraction (SciIE) is a vital task and is increasingly being adopted in biomedical data mining to conceptualize and epitomize knowledge triplets from the scientific literature. Existing relation extraction methods aim to extract explicit triplet knowledge from documents, however, they can hardly perceive unobserved factual relations. Recent generative methods have more flexibility, but their generated relations will encounter trustworthiness problems. In this paper, we first propose a novel Extraction-Contextualization-Derivation (ECD) strategy to generate a document-specific and entity-expanded dynamic graph from a shared static knowledge graph. Then, we propose a novel Dual-Graph Resonance Network (DGRN) which can generate richer explicit and implicit relations under the guidance of static and dynamic knowledge topologies. Experiments conducted on a public PubMed corpus validate the superiority of our method against several state-of-the-art baselines.

Accurately evaluating the impact of scientific papers is crucial. However, existing methodologies face certain challenges, including latent factors affecting citation behaviors and dynamic intrinsic of citation networks. To address these challenges, this study introduces a novel framework named CoDy (modeling scholarly Collaboration and temporal Dynamics in citation networks for impact prediction). CoDy strategically predicts author collaborations as an auxiliary task, forecasting not only the number of current collaborations between scholars but also the number of future collaborations among them. Besides, CoDy proposes a fine-grained temporal encoding module to model the multiple different temporal patterns for publication and citation. Extensive experimental validations demonstrate CoDy's effectiveness in predicting citation counts and classifying impact levels. In-depth analyses provide further validation of its reliability and robustness. CoDy can significantly enhance impact prediction by explicitly modeling collaboration and temporal patterns and offer valuable insights into paper impact formation.

User Satisfaction Estimation is an important task and increasingly being applied in goal-oriented dialogue systems to estimate whether the user is satisfied with the service. It is observed that whether the user’s needs are met often triggers various sentiments, which can be pertinent to the successful estimation of user satisfaction, and vice versa. Thus, User Satisfaction Estimation (USE) and Sentiment Analysis (SA) should be treated as a joint, collaborative effort, considering the strong connections between the sentiment states of speakers and the user satisfaction. Existing joint learning frameworks mainly unify the two highly pertinent tasks over cascade or shared-bottom implementations, however they fail to distinguish task-specific and common features, which will produce sub-optimal utterance representations for downstream tasks. In this paper, we propose a novel Speaker Turn-Aware Multi-Task Adversarial Network (STMAN) for dialogue-level USE and utterance-level SA. Specifically, we first introduce a multi-task adversarial strategy which trains a task discriminator to make utterance representation more task-specific, and then utilize a speaker-turn aware multi-task interaction strategy to extract the common features which are complementary to each task. Extensive experiments conducted on two real-world service dialogue datasets show that our model outperforms several state-of-the-art methods.

Xurui Li, Yue Qin, Rui Zhu, Tianqianjin Lin, Yongming Fan, Yangyang Kang, Kaisong Song, Fubang Zhao, Changlong Sun, Haixu Tang, Xiaozhong Liu. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Representation learning forms an essential building block in the development of natural language processing architectures.To date, mainstream approaches focus on learning textual information at the sentence-or document-level, unfortunately, overlooking the inter-document connections.This omission decreases the potency of downstream applications, particularly in multi-document settings.To address this issue, embeddings equipped with latent semantic and rich relatedness information are needed.In this paper, we propose SMRC 2 , which extends representation learning to the multi-document level.Our model jointly learns latent semantic information from content and rich relatedness information from topological networks.Unlike previous studies, our work takes multi-document as input and integrates both semantic and relatedness information using a shared space via language model and graph structure.Our extensive experiments confirm the superiority and effectiveness of our approach.To encourage further research in scientific multi-literature representation learning, we will release our code and a new dataset from the biomedical domain 1 .

Comparative Opinion Quintuple Extraction (COQE) is a task of recognizing comparative relationships in a sentence-level comment. It is additionally required to extract the quintuple constituents that attribute to a specific comparative relation, including a subject and object, as well as comparative aspect, opinion and preference. Previous study employ pipeline models in addressing the COQE task, which tend to suffer from error propagation. To address the issue, in this paper, we propose a BERT-based end-to-end neural model as the alternative. Specifically, we first boil COQE down to a set prediction problem, due to the finding that all the quintuple constituents fail to hold coherent or sequential relationships. In other word, we consider a group of quintuple components as a set, and intent to bag-and-drag them as a whole, instead individually and one-by-one. Furthermore, we leverage Graph Convolutional Network (GCN) to enhance the end-to-end model, which plays the role of perceiving and representing the relevant relations among the quintuple components. We experiment on three benchmark datasets, including Camera-COQE, Car-COQE and Ele-COQE. The experimental results show that our model (GCN-E2E) yields a significant improvement in most cases, compared to the BERT-based pipeline baseline. The performance reaches the F1-scores of 14.10%, 36.46% and 39.29%, with the improvements of 0.74%, 6.71% and 8.56%.

The sparseness and incompleteness of knowledge graphs (KGs) trigger considerable interest in enhancing the representation learning with external corpora. However, the difficulty of aligning entities and relations with external corpora leads to inferior performance improvement. Open knowledge graphs (OKGs) consist of entity-mentions and relation-mentions that are represented by noncanonicalized freeform phrases, which generally do not rely on the specification of ontology schema. The roughness of the nonontological construction method leads to a specific characteristic of OKGs: diversity, where multiple entity-mentions (or relation-mentions) have the same meaning but different expressions. The diversity of OKGs can provide potential textual and structural features for the representation learning of KGs. We speculate that leveraging OKGs to enhance the representation learning of KGs can be more effective than using pure text or pure structure corpora. In this paper, we propose a newOERL,Open knowledge graphEnhancedRepresentationLearning of KGs. OERL automatically extracts textual and structural connections between KGs and OKGs, models and transfers refined profitable features to enhance the representation learning of KGs. The strong performance improvement and exhaustive experimental analysis prove the superiority of OERL over state-of-the-art baselines.

Enterprise risk detection from financial documents (ERD) is known to be a key decision-making tool for a company that relies on the mass media. It would help those companies to prepare for potential risks in advance or prevent further deterioration of risks. However, ERD is usually hindered by its natural complexity and ambiguity. For a typical ERD task, the risk factors are considered to be multi-labeled while the inputs are mostly redundant. To overcome these difficulties, we proposed a hierarchical multi-task learning module ERD-NET. A novel article encoder is introduced to combine company information with the document’s relative information well. A hierarchical multi-task framework is also involved so that the main risk detection task could utilize the information learned from two easier auxiliary tasks. Our evaluation on the collected dataset shows that our proposed method outperformed the current state-of-art models.

Neural text ranking models have witnessed significant advancement and are increasingly being deployed in practice. Unfortunately, they also inherit adversarial vulnerabilities of general neural models, which have been detected but remain underexplored by prior studies. Moreover, the inherit adversarial vulnerabilities might be leveraged by blackhat SEO to defeat better-protected search engines. In this study, we propose an imitation adversarial attack on black-box neural passage ranking models. We first show that the target passage ranking model can be transparentized and imitated by enumerating critical queries/candidates and then train a ranking imitation model. Leveraging the ranking imitation model, we can elaborately manipulate the ranking results and transfer the manipulation attack to the target ranking model. For this purpose, we propose an innovative gradient-based attack method, empowered by the pairwise objective function, to generate adversarial triggers, which causes premeditated disorderliness with very few tokens. To equip the trigger camouflages, we add the next sentence prediction loss and the language model fluency constraint to the objective function. Experimental results on passage ranking demonstrate the effectiveness of the ranking imitation attack model and adversarial triggers against various SOTA neural ranking models. Furthermore, various mitigation analyses and human evaluation show the effectiveness of camouflages when facing potential mitigation approaches. To motivate other scholars to further investigate this novel and important problem, we make the experiment data and code publicly available.

We investigate response selection for multi-turn conversation in retrieval-based chatbots. Existing studies pay more attention to the matching between utterances and responses by calculating the matching score based on learned features, leading to insufficient model reasoning ability. In this paper, we propose a graph- reasoning network (GRN) to address the problem. GRN first conducts pre-training based on ALBERT using next utterance prediction and utterance order prediction tasks specifically devised for response selection. These two customized pre-training tasks can endow our model with the ability of capturing semantical and chronological dependency between utterances. We then fine-tune the model on an integrated network with sequence reasoning and graph reasoning structures. The sequence reasoning module conducts inference based on the highly summarized context vector of utterance-response pairs from the global perspective. The graph reasoning module conducts the reasoning on the utterance-level graph neural network from the local perspective. Experiments on two conversational reasoning datasets show that our model can dramatically outperform the strong baseline methods and can achieve performance which is close to human-level.

Identifying pornographic text online is practically useful to protect children from access to such adult content. However, some authors may intentionally avoid using sensitive words in their pornographic texts to take advantage of the lack of human audits. Without prior knowledge guidance, real semantics of such pornographic text is difficult to understand by existing methods due to its high context-sensitivity and heavy usage of figurative language, which brings huge challenges to the porn detection systems used in social media platforms. In this paper, we approach to the problem as a document-level porn identification task by locating and integrating sentence-level evidence and propose a novel Evidence-Aware Neural Porn Classification (eNPC) model. Specifically, we first propose a basic model which locates porn indicative sentences in the document with a multiple instance learning model, and then aggregate the sentence-level evidence to induce document label with self-attention mechanism. Moreover, we consider label dependencies within local context. Finally, we further enhance the sentence representation with prior knowledge produced by an automatic porn lexicon construction strategy. Extensive experimental results show that our model exhibits consistent superiority over competitors on two real-world Chinese novel datasets and an English story dataset.

Chatbot is increasingly thriving in different domains, however, because of unexpected discourse complexity and training data sparseness, its potential distrust hatches vital apprehension.Recently, Machine-Human Chatting Handoff (MHCH), predicting chatbot failure and enabling human-algorithm collaboration to enhance chatbot quality, has attracted increasing attention from industry and academia.In this study, we propose a novel model, Role-Selected Sharing Network (RSSN), which integrates both dialogue satisfaction estimation and handoff prediction in one multi-task learning framework.Unlike prior efforts in dialog mining, by utilizing local user satisfaction as a bridge, global satisfaction detector and handoff predictor can effectively exchange critical information.Specifically, we decouple the relation and interaction between the two tasks by the role information after the shared encoder.Extensive experiments on two public datasets demonstrate the effectiveness of our model.

Generating emotional responses is crucial for building human-like dialogue systems. However, existing studies have focused only on generating responses by controlling the agents' emotions, while the feelings of the users, which are the ultimate concern of a dialogue system, have been neglected. In this paper, we propose a novel variational model named EmoElicitor to generate appropriate responses that can elicit user's specific emotion. We incorporate the next-round utterance after the response into the posterior network to enrich the context, and we decompose single latent variable into several sequential ones to guide response generation with the help of a pre-trained language model. Extensive experiments conducted on real-world dataset show that EmoElicitor not only performs better than the baselines in term of diversity and semantic similarity, but also can elicit emotion with higher accuracy.

Kaisong Song, Lidong Bing, Wei Gao, Jun Lin, Lujun Zhao, Jiancheng Wang, Changlong Sun, Xiaozhong Liu, Qiong Zhang. 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.

Various types of target information have been considered in aspect-based sentiment analysis, such as entities and aspects. Existing research has realized the importance of targets and developed methods with the goal of precisely modeling their contexts via generating target-specific representations. However, all these methods ignore that these representations cannot be learned well due to the lack of sufficient human-annotated target-related reviews, which leads to the data sparsity challenge, a.k.a. cold-start problem here. In this paper, we focus on a more general multiple entity aspect-based sentiment analysis (ME-ABSA) task which aims at identifying the sentiment polarity of different aspects of multiple entities in their context. Faced with severe cold-start scenario, we develop a novel and extensible deep memory network framework with cold-start aware computational layers which use frequency-guided attention mechanism to accentuate on the most related targets, and then compose their representations into a complementary vector for enhancing the representations of cold-start entities and aspects. To verify the effectiveness of the framework, we instantiate it with a concrete context encoding method and then apply the model to the ME-ABSA task. Experimental results conducted on two public datasets demonstrate that the proposed approach outperforms state-of-the-art baselines on ME-ABSA task.

''User reviews” are becoming an essential component of e-commerce. When buyers write a negative or doubting review, ideally, the sellers need to quickly give a response to minimize the potential impact. When the number of reviews is growing at a frightening speed, there is an urgent need to build a response writing assistant for customer service providers. In order to generate high-quality responses, the algorithm needs to consume and understand the information from both the original review and the target product. The classical sequence-to-sequence (Seq2Seq) methods can hardly satisfy this requirement. In this study, we propose a novel deep neural network model based on the Seq2Seq framework for the review response generation task in e-commerce platforms, which can incorporate product information by a gated multi-source attention mechanism and a copy mechanism. Moreover, we employ a reinforcement learning technique to reduce the exposure bias problem. To evaluate the proposed model, we constructed a large-scale dataset from a popular e-commerce website, which contains product information. Empirical studies on both automatic evaluation metrics and human annotations show that the proposed model can generate informative and diverse responses, significantly outperforming state-of-the-art text generation models.

Emotion cause analysis has been a key topic in natural language processing. Existing methods ignore the contexts around the emotion word which can provide an emotion cause clue. Meanwhile, the clauses in a document play different roles on stimulating a certain emotion, depending on their content relevance. Therefore, we propose a co-attention neural network model for emotion cause analysis with emotional context awareness. The method encodes the clauses with a co-attention based bi-directional long short-term memory into high-level input representations, which are further fed into a convolutional layer for emotion cause analysis. Experimental results show that our approach outperforms the state-of-the-art baseline methods.

Review rating prediction is an important research topic. The problem was approached from either the perspective of recommender systems (RS) or that of sentiment analysis (SA). Recent SA research using deep neural networks (DNNs) has realized the importance of user and product interaction for better interpreting the sentiment of reviews. However, the complexity of DNN models in terms of the scale of parameters is very high, and the performance is not always satisfying especially when user-product interaction is sparse. In this paper, we propose a simple, extensible RS-based model, called Text-driven Latent Factor Model (TLFM), to capture the semantics of reviews, user preferences and product characteristics by jointly optimizing two components, a user-specific LFM and a product-specific LFM, each of which decomposes text into a specific low-dimension representation. Furthermore, we address the cold-start issue by developing a novel Pairwise Rating Comparison strategy (PRC), which utilizes the difference between ratings on common user/product as supplementary information to calibrate parameter estimation. Experiments conducted on IMDB and Yelp datasets validate the advantage of our approach over state-of-the-art baseline methods.

In the research of building emotion lexicons, we witness the exploitation of crowd-sourced affective annotation given by readers of online news articles. Such approach ignores the relationship between topics and emotion expressions which are often closely correlated. We build an emotion lexicon by developing a novel joint non-negative matrix factorization model which not only incorporates crowd-annotated emotion labels of articles but also generates the lexicon using the topic-specific matrices obtained from the factorization process. We evaluate our lexicon via emotion classification on both benchmark and built-in-house datasets. Results demonstrate the high-quality of our lexicon.