Wikipedia serves as the world's largest and most popular online reference encyclopedia, rich in structured knowledge and authoritative citations. Recently, numerous works have leveraged large language models to automatically generate Wikipedia-like articles. However, existing approaches primarily focus on producing singular narrative-type content, overlooking higher information-density structured elements such as timeline and table. To address these limitations, we propose WikiMAG, a multi-agent guided framework for generating structured Wikipedia-like articles. This framework employs a collaborative multi-agent mechanism to orchestrate the creation process, featuring three synergistic core components: Progressive planner first constructs the coarse-grained outline framework and then annotate fine-grained types for outline units, encompassing narrative, timeline, and table formats; Reflective inspector dynamically curates high-quality references via multi-round interactive feedback, thereby enhancing the authority and relevance of citations; Versatile writer integrates fine-grained outline details and high-quality reference information to generate information-rich articles, incorporating the three annotated formats. We evaluate WikiMAG on two public datasets, FreshWiki and WikiGenBen, across outline, writing, and verifiability dimensions. Compared with the best baseline method, our method achieves an average improvement of 6.73 points and 4.39 points in Heading Soft Recall and the METEOR metric (a machine translation and text generation evaluation metric) respectively, and an average increase of 16.84 percentage points in Citation Rate.

High-quality multi-hop instruction data is critical for enhancing the reasoning capabilities of large language models (LLMs) in complex long-context scenarios, e.g., long-form reasoning. Nevertheless, there is currently a notable scarcity of such datasets within the community, and existing data synthesis approaches typically fail to provide explicit modeling of intermediate reasoning steps, resulting in unverifiable and potentially erroneous samples. To mitigate above issue, we design the Concept-Graph based Multi-hop Instructions Synthesis (CGMIS) framework, which constructs long-form reasoning paths via concept graph traversal and automatically generates verifiable multi-hop data. The CGMIS framework not only guarantees the accuracy and verifiability of the synthesized data but also enables the construction of high-quality multi-hop instruction datasets from arbitrary corpora. Experiments show that fine-tuning with CGMIS-generated data achieves state-of-the-art performance across 13 long-context reasoning tasks on various models, using only 10% of the data volume required by existing methods.

Long-form books are among the most information-rich and structurally complex forms of written content, often exceeding 100,000 words. While recent methods have enabled basic long-text generation, they remain limited in two key aspects: the inability to generate ultra-long content at book scale, and the lack of mechanisms for integrating rich factual information. To address these limitations, we propose DeepWriter, a multi-agent collaborative framework that follows a structured planning-then-generation paradigm. It first constructs a detailed book outline with narrative arcs and chapter semantics, then incrementally generates content conditioned on retrieved knowledge and contextual signals. DeepWriter supports controllable generation of full-length books exceeding 100,000 words, enriched with citations, trivia and images. To support evaluation beyond surface-level fluency, we introduce DeepWriter-Bench, a bilingual benchmark of 18 annotated books designed to assess book-scale coherence, richness, and factual grounding. Additionally, we propose BookScore, a unified 100-point metric for quantifying book maturity. Experimental results show that DeepWriter achieves a state-of-the-art BookScore of 80.92, consistently outperforming strong baselines.

Large Reasoning Models (LRMs) achieve promising results on complex reasoning tasks but remain susceptible to hallucinations. Existing hallucination detection methods based on Large Language Models (LLMs) often focus solely on final answers, overlooking inconsistencies between the answer and reasoning process. This limitation reduces their ability to detect hallucinations during inference. Moreover, training-free approaches lack mechanisms for confidence estimation, resulting in an unquantified detection output. In contrast, training-based methods can provide fine-grained assessments but often neglect the self-correction capability of LRMs, where earlier errors may be corrected in subsequent steps, leading to inaccurate hallucination detection. To address these challenges, we propose ConfFuse, a unified framework that fuses global and local confidence scores for hallucination detection. A Global Hallucination Detection Model (GHDM) is trained using Direct Preference Optimization (DPO) to assess hallucinations at the level of entire reasoning chains, yielding global confidence estimates. Simultaneously, a Process Reward Model (PRM) estimates step-wise confidence scores to capture local logical flaws. A weighted fusion strategy combines the global confidence score with the minimum local score to jointly reflect overall reasoning consistency and local soundness. Experimental evaluations demonstrate that ConfFuse surpasses Qwen3-1.7B and Qwen3-8B by up to 11.86% and 5.46% in F1 score on in-distribution datasets, and achieves average improvements of 4.65% and 2.80% on out-of-distribution datasets. These results verify the effectiveness and generalizability of the proposed framework.

Zechen Sun, Yuyang Sun, Zecheng Tang, Juntao Li, Wenpeng Hu, Wenliang Chen, Zhunchen Luo, Guotong Geng, Min Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Bo Zhang, Jiawei Zhang, Cong Gao, Bingxu Han, Minghao Hu, Jun Zhang, Yunbo Cao, Zhunchen Luo, Wen Yao, Guotong Geng, Zhong Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Multimodal Entity Linking (MEL) aims to associate textual and visual mentions with entities in a multimodal knowledge graph. Despite its importance, current methods face challenges such as incomplete contextual information, coarse cross-modal fusion, and the difficulty of jointly large language models (LLMs) and large visual models (LVMs). To address these issues, we propose DeepMEL, a novel framework based on multi-agent collaborative reasoning, which achieves efficient alignment and disambiguation of textual and visual modalities through a role-specialized division strategy. DeepMEL integrates four specialized agents, namely Modal-Fuser, Candidate-Adapter, Entity-Clozer and Role-Orchestrator, to complete end-to-end cross-modal linking through specialized roles and dynamic coordination. DeepMEL adopts a dual-modal alignment path, and combines the fine-grained text semantics generated by the LLM with the structured image representation extracted by the LVM, significantly narrowing the modal gap. We design an adaptive iteration strategy, combines tool-based retrieval and semantic reasoning capabilities to dynamically optimize the candidate set and balance recall and precision. DeepMEL also unifies MEL tasks into a structured cloze prompt to reduce parsing complexity and enhance semantic comprehension. Extensive experiments on five public benchmark datasets demonstrate that DeepMEL achieves state-of-the-art performance, improving ACC by 1 %-57 %. Ablation studies verify the effectiveness of all modules.

BERT-family have been increasingly explored for adaptation to scenarios beyond language understanding tasks, with more recent efforts focused on enabling them to become good instruction followers.These explorations have endowed BERT-family with new roles and human expectations, showcasing their potential on par with current state-of-the-art (SOTA) large language models (LLMs).However, several certain shortcomings in previous BERT-family, such as the relatively sub-optimal training corpora, learning procedure, and model architecture, all impede the further advancement of these models for serving as general and competitive LLMs.Therefore, we aim to address these deficiencies in this paper.Our study not only introduces a more suitable pre-training task that helps BERT-family excel in wider applications to realize generality but also explores the integration of cutting-edge technologies into our model to further enhance their capabilities.Our final models, termed Bidirectional General Language Models (BiGLM), exhibit performance levels comparable to current SOTA LLMs across a spectrum of tasks.Moreover, we conduct detailed analyses to study the effects of scaling and training corpora for BiGLM.To the best of our knowledge, our work represents the early attempt to offer a recipe for building novel types of scalable, general, and competitive LLMs that diverge from current autoregressive modeling methodology.Our codes and models are available on Github 1 .

Adapting Multi-modal Large Language Models (MLLMs) to target tasks often suffers from catastrophic forgetting, where acquiring new task-specific knowledge compromises performance on pre-trained tasks. In this paper, we introduce AdaDARE-γ, an efficient approach that alleviates catastrophic forgetting by controllably injecting new task-specific knowledge through adaptive parameter selection from fine-tuned models without requiring retraining procedures. This approach consists two key innovations: (1) an adaptive parameter selection mechanism that identifies and retains the most task-relevant parameters from fine-tuned models, and (2) a controlled task-specific information injection strategy that precisely balances the preservation of pre-trained knowledge with the acquisition of new capabilities. Theoretical analysis proves the optimality of our parameter selection strategy and establishes bounds for the task-specific information injection factor. Extensive experiments on InstructBLIP and LLaVA-1.5 across image captioning and visual question answering tasks demonstrate that AdaDARE-γ establishes new state-of-the-art results in balancing model performance. Specifically, it maintains 98.2% of pre-training effectiveness on original tasks while achieving 98.7% of standard fine-tuning performance on target tasks.

Understanding the underlying argumentative flow in analytic argumentative writing is essential for discourse comprehension, especially in complex argumentative discourse such as think-tank commentary.However, existing structure modeling approaches often rely on surface-level topic segmentation, failing to capture the author's rhetorical intent and reasoning process.To address this limitation, we propose a Question-Focus discourse structuring framework that explicitly models the underlying argumentative flow by anchoring each argumentative unit to a guiding question (reflecting the author's intent) and a set of attentional foci (highlighting analytical pathways).To assess its effectiveness, we introduce an argument reconstruction task in which the modeled discourse structure guides both evidence retrieval and argument generation.We construct a high-quality dataset comprising 600 authoritative Chinese think-tank articles for experimental analysis.To quantitatively evaluate performance, we propose two novel metrics: (1) Claim Coverage, measuring the proportion of original claims preserved or similarly expressed in reconstructions, and (2) Evidence Coverage, assessing the completeness of retrieved supporting evidence.Experimental results show that our framework uncovers the author's argumentative logic more effectively and offers better structural guidance for reconstruction, yielding up to a 10% gain in claim coverage and outperforming strong baselines across both curated and LLM-based metrics.

Various research has addressed the overconfidence problem, and we focus on improving the robustness of post-hoc calibration (e.g., temperature scaling, TS) when the test set shifts from the training set by image corruption. TS is greatly affected by the validation set, which previous work has proposed to perturb by Gaussian noise to improve calibration under domain drift. Inspired by this, we discovered that the same or similar augmentation on the validation set substantially improved TS under corrupted shift. We proposed a learnable and dynamic augmentation-based TS method, AugTS, which minimizes the maximum mean discrepancy (MMD) between the augmented validation and corrupted test set. Experiments on corrupted versions of CIFAR-10, CIFAR-100, and TinyImageNet show that AugTS can significantly improve calibration under corrupted shifts compared with competitive baselines.

This paper addresses the challenge of insufficient sampling diversity in large language models (LLMs) under the conventional autoregressive decoding framework. Our research reveals that, in some tasks where LLMs underperform, they actually possess the capability to provide correct answers. However, this potential is limited by inadequate sampling diversity in the decoding process, which prevents the models from effectively and fully leveraging their reasoning capabilities. To address this issue, we propose the Adaptive Trigram Model-Assisted Decoding Strategy (ATM-ADS), a method designed to enhance the reasoning abilities of large models without requiring fine-tuning. By dynamically providing diverse candidate vocabularies, combined with a decoding dynamic decision-making mechanism and an adaptive smoothing strategy, our approach substantially enhances sampling diversity during the decoding process. Experimental results demonstrate that our ATM-ADS significantly improves model performance across multiple task scenarios, highlighting its broad potential for cross-task and cross-linguistic applications.

Relation triplet extraction aims to identify entity pairs and their relations from unstructured text. Traditional zero-shot learning methods are hindered by predefined relation types and the lack of large-scale annotated data, limiting generalization to unseen relations. To address these challenges, we propose the KOMADF (Knowledge-Optimized Multi-Agent Dynamics Framework). KOMADF constructs task-specific knowledge graphs and incorporates a distillation module to refine relation labels, filtering out irrelevant candidates while preserving semantically similar or multi-type labels. These refined labels generate problem representations that guide a multi-agent mechanism to dynamically create context-aware prompt templates. By aligning task objectives with pre-trained language models, these templates enable generalization to unseen relations.Experiments on two public datasets, FewRel and Wiki-ZSL, show that the proposed method substantially improves precision, recall, and F1 score. These findings confirm KOMADF’s adaptability and robustness, establishing it as an effective solution to the challenges of zero-shot relation triplet extraction. Furthermore, the proposed framework lays a solid theoretical foundation for the automated construction of dynamic knowledge graphs.

This study explores the efficacy of fine-tuning large language models (LLMs) using non-task-specific datasets, challenging the traditional reliance on task-specific datasets. Based on the SUP-NATINST and Stanford Alpaca dataset, this research explores some schemes to enhance the application capabilities of LLMs in specific domains through strategic data selection. The main contributions include an innovative fine-tuning data selection strategy that emphasizes the importance of dataset selection by comparing non-task-specific datasets with task-specific datasets, demonstrating the potential advantages of unconventional datasets in improving model task performance. The study also reveals the model’s performance sensitivity to data ratios, challenges the concept of optimal data ratios, and explores the impact of data volume changes on model accuracy. Furthermore, it deepens the understanding of how different task types affect model performance fine-tuning. By analyzing the performance across different tasks and dimensions and dissecting the impact of data ratio changes, the research indicates that non-specific task data allocation can achieve better performance than specific task datasets.

Large language models (LLMs) have demonstrated the ability to perform in-context learning (ICL) with only a few demostrations, achieving remarkable performance across various downstream tasks. The selection of demonstrations plays a critical role in shaping the performance of ICL due to its high sensitivity. However, previous researchers have primarily focused on either the tricks for selecting demonstrations or the sequencing of demonstrations, thereby neglecting the critical role of retrieval models in ICL. The rigid structures and parameters employed in these studies often fail to align with the specific requirements of downstream tasks. To address this problem, we propose CLUG: contrastive learning unified retrieval with graph-ranked demonstrations for enhanced in-context learning, integrating demonstrations retrieval selection and demonstrations order to establish semantically coherent sequences of demonstrations, thereby ensuring enhanced semantic alignment and consistency. Experimental results across multiple datasets demonstrate consistent improvements with our method. Additional analyses further validate and explain the effectiveness and generalizability of our approach.

Event argument extraction (EAE) is an important and critical task in natural language processing, whose goal is to transform unstructured text into structured information. However, previous methods are costly for annotation, which limits their application. With the emergence of large language models (LLMs), many approaches use prompt learning or instruction learning to solve the EAE task. However, due to the complexity of the EAE, the defined instructions are often abstract and difficult for the model to accurately understand the task objective, while the cost of gradient propagation based methods is too high. Therefore, we propose an Instruction Feature Adaptation event argument extraction (IFA) that uses very few shots to tune the task instruction features, while not requiring a highcost gradient optimisation method to enhance the model’s performance on the event argument extraction task. Experiments demonstrate that the method is able to optimise the instruction features, enabling the model to perform the event argument task more consistently based on the instructions, ultimately improving performance for low-resource. Code is available at https://github.com/yangzonghao1024/IFA.

A scientific claim typically begins with the formulation of a research question or hypothesis, which is a tentative statement or proposition about a phenomenon or relationship between variables.Within the realm of scientific claim verification, considerable research efforts have been dedicated to attention architectures and leveraging the text comprehension capabilities of Pre-trained Language Models (PLMs), yielding promising performances.However, these models overlook the causal structure information inherent in scientific claims, thereby failing to establish a comprehensive chain of causal inference.This paper delves into the exploration to highlight the crucial role of qualitative causal structure in characterizing and verifying scientific claims based on evidence.We organize the qualitative causal structure into a heterogeneous graph and propose a novel attentionbased graph neural network model to facilitate causal reasoning across relevant causallypotent factors.Our experiments demonstrate that by solely utilizing the qualitative causal structure, the proposed model achieves comparable performance to PLM-based models.Furthermore, by incorporating semantic features, our model outperforms state-of-the-art approaches comprehensively. 1

In this paper, we study the problem of identifying the principals and accessories from the fact description with multiple defendants in a criminal case.We treat the fact descriptions as narrative texts and the defendants as roles over the narrative story.We propose to model the defendants with behavioral semantic information and statistical characteristics, then learning the importances of defendants within a learning-to-rank framework.Experimental results on a real-world dataset demonstrate the behavior analysis can effectively model the defendants' impacts in a complex case.

He Zhao, Zhunchen Luo, Chong Feng, Anqing Zheng, Xiaopeng Liu. 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.

With the growth in popularity and size of social media, there is an urgent need for systems that can recognize the coreference relation between two event mentions in texts from social media. In existing event coreference resolution research, a rich set of linguistic features derived from pre-existing NLP tools and various knowledge bases is often required. This kind of methods restricts domain scalability and leads to the propagation of errors. In this paper, we present a novel selective expression approach based on event trigger to explore the coreferential relationship in high-volume Twitter texts. Firstly, we exploit a bidirectional Long Short Term Memory (Bi-LSTM) to extract the sentence level and mention level features. Then, to selectively express the essential parts of generated features, we apply a gate on sentence level features. Next, to integrate the time information of event mention pairs, we design an auxiliary feature based on triggers and time attributes of the two event mentions. Finally, all these features are concatenated and fed into a classifier to predict the binary coreference relationship between the event mention pair. To evaluate our method, we publish a new dataset EventCoreOnTweet (ECT)1that annotates the coreferential relationship between event mentions and event trigger of each event mention. The experimental results demonstrate that our approach achieves significant performance in the ECT dataset.

In this paper, we introduce the task of resource citation classification for scientific literature using a context-based framework. This task is to analyze the purpose of citing an on-line resource in scientific text by modeling the role and function of each resource citation. It can be incorporated into resource indexing and recommendation systems to help better understand and classify on-line resources in scientific literature. We propose a new annotation scheme for this task and develop a dataset of 3,088 manually annotated resource citations. We adopt a neural-based model to build the classifiers and apply them on the large ARC dataset to examine the revolution of scientific resources from trends in their function over time.

Charge prediction which aims to determine appropriate charges for criminal cases based on textual fact descriptions, is an important technology in the field of AI&Law. Previous works focus on improving prediction accuracy, ignoring the interpretability, which limits the methods’ applicability. In this work, we propose a deep neural framework to extract short but charge-decisive text snippets – rationales – from input fact description, as the interpretation of charge prediction. To solve the scarcity problem ofrationale annotatedcorpus, rationalesare extractedinareinforcement stylewiththe only supervision in the form of charge labels. We further propose a dynamic rationale attention mechanism to better utilize the information in extracted rationales and predict the charges. Experimental results show that besides providing charge prediction interpretation, our approach can also capture subtle details to help charge prediction.

We present Proteus, a novel network-aware approach for optimizing web browsing on heterogeneous multi-core mobile systems. It employs machine learning techniques to predict which of the heterogeneous cores to use to render a given webpage and the operating frequencies of the processors. It achieves this by first learning offline a set of predictive models for a range of typical networking environments. A learnt model is then chosen at runtime to predict the optimal processor configuration, based on the web content, the network status and the optimization goal. We evaluate Proteus by implementing it into the open-source Chromium browser and testing it on two representative ARM big.LITTLE mobile multi-core platforms. We apply Proteus to the top 1,000 popular websites across seven typical network environments. Proteus achieves over 80% of best available performance. It obtains, on average, over 17% (up to 63%), 31% (up to 88%), and 30% (up to 91%) improvement respectively for load time, energy consumption and the energy delay product, when compared to two state-of-the-art approaches.

Event extraction is of practical utility in natural language processing.In the real world, it is a common phenomenon that multiple events existing in the same sentence, where extracting them are more difficult than extracting a single event.Previous works on modeling the associations between events by sequential modeling methods suffer a lot from the low efficiency in capturing very long-range dependencies.In this paper, we propose a novel Jointly Multiple Events Extraction (JMEE) framework to jointly extract multiple event triggers and arguments by introducing syntactic shortcut arcs to enhance information flow and attention-based graph convolution networks to model graph information.The experiment results demonstrate that our proposed framework achieves competitive results compared with state-of-the-art methods.

Hai Ye, Xin Jiang, Zhunchen Luo, Wenhan Chao. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Connections between relations in relation extraction, which we call class ties, are common.In distantly supervised scenario, one entity tuple may have multiple relation facts.Exploiting class ties between relations of one entity tuple will be promising for distantly supervised relation extraction.However, previous models are not effective or ignore to model this property.In this work, to effectively leverage class ties, we propose to make joint relation extraction with a unified model that integrates convolutional neural network (CNN) with a general pairwise ranking framework, in which three novel ranking loss functions are introduced.Additionally, an effective method is presented to relieve the severe class imbalance problem from NR (not relation) for model training.Experiments on a widely used dataset show that leveraging class ties will enhance extraction and demonstrate the effectiveness of our model to learn class ties.Our model outperforms the baselines significantly, achieving stateof-the-art performance.

Spam and wildly varying documents make searching in Twitter challenging. Most Twitter search systems generally treat a Tweet as a plain text when modeling relevance. However, a series of conventions allows users to Tweet in structural ways using a combination of different blocks of texts. These blocks include plain texts, hashtags, links, mentions, etc. Each block encodes a variety of communicative intent and the sequence of these blocks captures changing discourse. Previous work shows that exploiting the structural information can improve the structured documents (e.g., web pages) retrieval. In this study we utilize the structure of Tweets, induced by these blocks, for Twitter retrieval and Twitter opinion retrieval. For Twitter retrieval, a set of features, derived from the blocks of text and their combinations, is used into a learning‐to‐rank scenario. We show that structuring Tweets can achieve state‐of‐the‐art performance. Our approach does not rely on social media features, but when we do add this additional information, performance improves significantly. For Twitter opinion retrieval, we explore the question of whether structural information derived from the body of Tweets and opinionatedness ratings of Tweets can improve performance. Experimental results show that retrieval using a novel unsupervised opinionatedness feature based on structuring Tweets achieves comparable performance with a supervised method using manually tagged Tweets. Topic‐related specific structured Tweet sets are shown to help with query‐dependent opinion retrieval.

An important aspect of communication in Twitter (and other Social Network is message propagation -- people creating posts for others to share. Although there has been work on modelling how tweets in Twitter are propagated (retweeted), an untackled problem has been who will retweet a message. Here we consider the task of finding who will retweet a message posted on Twitter. Within a learning to-rank framework, we explore a wide range of features, such as retweet history, followers status, followers active time and followers interests. We find that followers who retweeted or mentioned the author's tweets frequently before and have common interests are more likely to be retweeters.

Most Twitter search systems generally treat a tweet as a plain text when modeling relevance. However, a series of conventions allows users to tweet in structural ways using combination of different blocks of texts.These blocks include plain texts, hashtags, links, mentions, etc. Each block encodes a variety of communicative intent and sequence of these blocks captures changing discourse. Previous work shows that exploiting the structural information can improve the structured document (e.g., web pages) retrieval. In this paper we utilize the structure of tweets, induced by these blocks, for Twitter retrieval. A set of features, derived from the blocks of text and their combinations, is used into a learning-to-rank scenario. We show that structuring tweets can achieve state-of-the-art performance. Our approach does not rely upon social media features, but when we do add this additional information, performance improves significantly.