Large language model (LLM) generated texts now rival human quality, creating four text categories: purely machine-generated, machine-rewritten, machine-polished, and human-written content. Traditional detection methods face significant challenges in human-machine hybrid scenarios where LLMs perform rewriting or polishing, as existing approaches focus on single-level features and fail to capture subtle, multi-layered machine traces. To address this, we propose the Multi-level Style Preference Optimization (MSPO) framework, capturing machine style features at multiple granularities: sequence-level (overall consistency), phrase-level (distinctive n-gram patterns), and lexical-level (word selection distributions). We further incorporate four text complexity indicators (Type-Token Ratio, Average Sentence Length, Average Word Length, and Punctuation Ratio) to dynamically adjust optimization parameters based on human-machine text complexity differences, enhancing adaptability across diverse text types. Additionally, we construct a comprehensive detection dataset spanning three representative domains (scientific writing, news articles, and creative writing) across four text types (human-written, purely machine-generated, machine-rewritten, and machine-polished), generated using state-of-the-art LLMs for robust evaluation. Experimental results demonstrate that MSPO significantly outperforms existing methods across all text types. On the challenging rewritten texts, MSPO achieves up to 82.14% AUROC, representing an improvement of 11.15 percentage points over the strongest baseline ImBD, while maintaining robust cross-domain generalizability across scientific, news, and creative writing domains.

Current text style transfer task mainly focuses on short texts, while the field has not been fully developed for long texts. Considering the richer semantics and more complex sentence structures in long text sequences, existing methods that employ traditional style-content disentanglement ways and learn the target style to generate target sequences face two key issues: 1) During disentanglement, they usually directly separate style words or fragments, such coarse-grained disentanglement risks losing original semantics and hinder the model's content preservation. 2) During target style learning, they often focus on the transfer of certain style attributes or aspects, which makes it difficult to grasp the holistic style of target objects. To this end, we propose Cognitive enhancement Chain-of-Thought (CeCoT) towards enhancing style learning and content preservation for long style transfer. CeCoT first constructs progressive CoT to facilitate LLMs to gradually rewrite source content and separate source styles, thereby enhancing the retention of original content. Then, we propose cognitive CoT, which comprehensively considers hierarchical cognitive content (i.e., shallower-deeper-normal level) and cognitive behavior (i.e., prompt order of CoT) to learn the overall target style. To enhance the robustness of our model, we also propose two constraint losses in a dual validation way towards content preservation enhancing and style consistency learning. Extensive experiments on two competitive datasets demonstrate the superiority of our CeCoT.

Empathetic dialogue requires not only recognizing a user's emotional state but also making strategy-aware, context-sensitive decisions throughout response generation.However, the lack of a comprehensive empathy strategy framework, explicit task-aligned multi-stage reasoning, and high-quality strategy-aware data fundamentally limits existing approaches, preventing them from effectively modeling empathetic dialogue as a complex, multi-stage cognitive and decision-making process.To address these challenges, we propose STRIDE-ED, a STRategy-grounded, Interpretable, and DEep reasoning framework that models Empathetic Dialogue through structured, strategyconditioned reasoning.To support effective learning, we develop a strategy-aware data refinement pipeline integrating LLM-based annotation, multi-model consistency-weighted evaluation, and dynamic sampling to construct highquality training data aligned with empathetic strategies.Furthermore, we adopt a two-stage training paradigm that combines supervised fine-tuning with multi-objective reinforcement learning to better align model behaviors with target emotions, empathetic strategies, and response formats.Extensive experiments demonstrate that STRIDE-ED generalizes across diverse open-source LLMs and consistently outperforms existing methods on both automatic metrics and human evaluations.

Jingchun Lian, Lingyu Liu, Yaxiong Wang, Yujiao Wu, Lianwei Wu, Li Zhu, Zhedong Zheng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

The rapid growth of short video platforms has made multimodal fake news more prevalent.Existing detectors suffer from two major limitations: (I) global-alignment bias that overemphasizes holistic cross-modal matching and thus misses subtle, localized inconsistencies; and (II) LLM-based methods that leverage powerful generative reasoning to identify cognitive forgeries but inherently suffer from hallucinations and high inference latency.To overcome these limitations, we propose PCDD, a novel Perception-Cognition Dualdriven Detector that jointly observes the form and probes the logic for short video fake news detection.The perception stream exposes finegrained cross-modal conflicts by amplifying localized inconsistencies into explicit discrepancies.The cognition stream transfers reasoning capabilities from LLMs to a lightweight student to mine cognitive forgeries, while reducing the risk of hallucinations and eliminating reliance on LLMs at inference.Experiments on real-world datasets show that PCDD consistently outperforms baselines, while improving interpretability and robustness in data scarcity scenarios.

Yuchen Zhang, Yaxiong Wang, Kecheng Han, Yujiao Wu, Lianwei Wu, Li Zhu, Zhedong Zheng. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

This paper studies the network dismantling (ND) problem and aims to develop more effective models and approaches to cope with it, such that a given network can be dismantled by a set of vital nodes of minimum size. To achieve that, we propose a three-phase framework—the Percolation coarsening, Belief propagation dismantling, and Fragmentation optimization fine-tuning (PBF) framework—consisting of PBF-I, PBF-II, and PBF-III, where we contribute three new and one improved algorithms. In particular, PBF-I studies strategies to effectively coarsen the studied network via the merger of less influential nodes, such that the computational efficiency of the follow-up PBF-II phase can be maximized. PBF-II considers the superiority of the belief propagation (BP) algorithm in the ND problem and proposes an improved BP to identify vital nodes from the coarse-grained network, which particularly focuses on the largest connected component and obtains the vital nodes from a filtered candidate set. In addition, PBF-III presents fine-tuning strategies to further improve the quality of solutions obtained in PBF-II. The effectiveness of the proposed framework is validated on over 10 empirical networks in regard to varied circumstances. Our results show that the developed framework can obtain dismantling node sets of much smaller sizes compared to the state-of-the-art in almost all cases. Meanwhile, our framework is also more effective, efficient, and stable compared to existing methods, and is capable of tackling the ND problem in extremely large networks. We are convinced that the model and methodology introduced in this paper could be applied to many applications, such as the robustness and resilience analysis of network-structural infrastructures, the suppression of epidemics, and the containment of misinformation on social networks. The source code of the proposed PBF framework will be made publicly available upon acceptance of the manuscript.

Large Language Models (LLMs) have permeated various Natural Language Processing (NLP) tasks. For the summarization tasks, LLMs can generate well-structured rationales, which consist of Essential Aspects (EA), Associated Sentences (AS) and Triple Entity Relations (TER). These rationales guide smaller models (≤1B) to produce better summaries. However, their high deployment costs (≥70B), such as substantial storage space and high computing requirements, limit their utilization in resource-constrained environments. Furthermore, effectively distilling these structured rationales from LLMs into Small Language Models (SLMs) models remains a challenge. To address this, we propose the LLM-based Structured Rationale-guided Multi-view Weak-gated Fusion framework (LSR-MWF). The framework initially employs LLMs to dig structural rationales from a document, considering multiple viewpoints such as EA, AS, and TER. Then, it develop a multi-step summary generation evaluation strategy to select high-quality structured rationales. Subsequently, it aligns with these rationales using additional modules organized in a hierarchical structure. Finally, the framework integrates the features output by these modules with original abstractive model through a weak-gated mechanism. Experimental results on two publicly available CNN/DailyMail and XSum datasets show that our method improves the performance of the abstractive model, outperforming baselines by 11.2％ and 5.8％, respectively. In addition, our method improves the interpretability of summary generation from the viewpoints of EA, AS and TER.

Multi-modal large language models (MLLMs) are expanded from large language models (LLMs) with additional capabilities to infer multi-modal data. Current MLLM workflows, when dealing with complex tasks, typically begin by using an LLM to decompose the task into multiple subtasks, then heuristically select a specific pre-trained model to complete a subtask to get a result, and finally integrate all the results to obtain the final response. However, heuristically binding one model to one subtask may generate a less satisfying subtask result, thereby affecting the overall performance. Therefore, we propose CompeMLLM, which introduces an innovative method of dynamic orchestration of the workflows. It allows different models to compete on the same subtask instead of statically binding them. By dynamically integrating the results from diverse models, the optimal subtask result is determined, thereby improving the overall performance of MLLM. Specifically, given a certain complex task, CompeMLLM first decomposes it into subtasks, then employs multiple pre-trained models to execute the same subtask in parallel to compete, and then the optimal subtask result is chosen by dynamically evaluating these results using ensemble learning idea, and finally integrates these optimal results into a complete workflow, thus obtaining the best overall performance. We conducted extensive experiments using six advanced MLLMs as baselines across seven diverse datasets. The experimental results robustly demonstrate that CompeMLLM achieves significantly improved performance on all datasets, demonstrating its effectiveness.

Graph generation is essential in social network analysis, particularly for modeling information flow and user interactions. However, existing probabilistic diffusion models face challenges when applied to social propagation graphs. The continuous noise does not apply to the discrete nature of graph generation tasks, and the random Gaussian initialization in the reverse process can introduce biases that deviate from real-world propagation patterns. To address these issues, this paper introduces a Prior-based Discrete Diffusion Model (PDDM) for social graph generation. PDDM redefines the forward process as a discrete process for node denoising and edge generation, and the task of the denoising module is transformed into the connection probability learning of node-level tasks. Further, PDDM employs a new starting point of the reverse process by incorporating user similarity as the probability matrix, which can better leverage the social context. These developments mitigate reverse-starting bias and enhance model robustness. Moreover, PDDM integrates lightweight deep graph networks such as GAT, demonstrating both scalability and applicability to graph generation scenarios. Comprehensive experiments on real-world social network datasets demonstrate PDDM’s superiority in terms of the MMD metric and downstream tasks. The code is available at https://github.com/cgao-comp/PDDM.

With the maturation and advancement of large language models (LLMs), personalizing content generation has emerged as a significant challenge. Traditional approaches rely on retrieving historical data from users with similar queries and either retraining small language models or fine-tuning large models for specific tasks, both requiring substantial computational resources and maintenance. To take advantage of the increasing capabilities and cost-effectiveness of publicly deployed LLMs, we propose the Personalized Retrieval-Evolution Framework (PREF). This framework innovatively combines historical data retrieval with cost-effective LLM API calls as evaluation functions through strategic prompt reuse. Additionally, it employs genetic algorithms to identify keywords that best characterize user styles and characteristics. Our approach surpasses conventional methods that combine fine-tuned smaller models with historical data retrieval, while eliminating the computational overhead associated with repeated model training during task transitions. The framework demonstrates a more efficient and adaptable solution for personalized content generation in the era of large language models.

Recent progress in aligning images with texts has achieved remarkable results, however, existing models tend to serve general queries and often fall short when dealing with detailed query requirements. In this paper, we work towards Entity-centric Image-Text Matching (EITM), a finer-grained image-text matching task that aligns texts and images centered around specific entities. The main challenge in EITM lies in bridging the substantial semantic gap between entity-related information in texts and images, which is more pronounced than in general image-text matching problems. To address this challenge, we adopt CLIP as our foundational model and devise a Multimodal Attentive Experts (MMAE)-based contrastive learning to adapt CLIP into an expert for EITM problem. Particularly, the core of our multimodal attentive experts learning is to generate explanation texts by Large Language Models (LLMs) as bridging clues. In specific, we first employ off-the-shelf LLMs to generate explanatory text. This text, along with the original image and text, is then fed into our Multimodal Attentive Experts module to narrow the semantic gap within a unified semantic space. Upon the enriched feature representations generated by MMAE, we have further developed an effective Gated Integrative Image-text Matching (GI-ITM) strategy. GI-ITM utilizes an adaptive gating mechanism to combine features from MMAE, followed by applying image-text matching constraints to enhance the alignment precision. Our method has been extensively evaluated on three social media news benchmarks: N24News, VisualNews, and GoodNews. The experimental results demonstrate that our approach significantly outperforms competing methods. Our code is available at: https://github.com/wangyxxjtu/ETE.

Previous studies on multimodal fake news detection mainly focus on the alignment and integration of cross-modal features, as well as the application of text-image consistency. However, they overlook the semantic enhancement effects of large multimodal models and pay little attention to the emotional features of news. In addition, people find that fake news is more inclined to contain negative emotions than real ones. Therefore, we propose a novel Semantic Enhancement and Emotional Reasoning (SEER) Network for multimodal fake news detection. We generate summarized captions for image semantic understanding and utilize the products of large multimodal models for semantic enhancement. Inspired by the perceived relationship between news authenticity and emotional tendencies, we propose an expert emotional reasoning module that simulates real-life scenarios to optimize emotional features and infer the authenticity of news. Extensive experiments on two real-world datasets demonstrate the superiority of our SEER over state-of-the-art baselines.

Fact verification task has emerged as an essential research topic recently due to abundant fake news spreading on the Internet. The task based on unstructured data (i.e., news) has achieved great development, but the task based on structured data (i.e., table) is still in the primary development period. The existing methods usually construct complete heterogeneous graph networks around statement, table, and program subgraphs, and then infer to learn similar semantics on them for fact verification. However, they generally connect the nodes with the same content between subgraphs directly to frame a larger graph network, which has serious sparsity in connections, especially when subgraphs possess limited semantics. To this end, we propose tight-fitting graph inference network (TFGIN), which innovatively builds tight-fitting graphs (TF-graphs) to strengthen the connections of subgraphs and designs inference modeling layer (IML) to learn coherence evidence for fact verification. Specifically, different from traditional connection ways, the constructed TF-graph enhances inter-graph and intra-graph connections of subgraphs through subgraph segmentation and interaction guidance mechanisms. IML could reason the semantics with strong correlation and high consistency as explainable evidence. Experiments on three competitive datasets confirm the superiority and scalability of our TFGIN.

With the rise of social media, the spread of fake news has become a significant concern, potentially misleading public perceptions and impacting social stability. Although deep learning methods like CNNs, RNNs, and Transformer-based models like BERT have enhanced fake news detection. However, they primarily focus on content and do not consider social context during news propagation. Graph-based techniques have incorporated the social context but are limited by the need for large labeled datasets. To address these challenges, this paper introduces GAMC, an unsupervised fake news detection technique using the Graph Autoencoder with Masking and Contrastive learning. By leveraging both the context and content of news propagation as self-supervised signals, our method reduces the dependency on labeled datasets. Specifically, GAMC begins by applying data augmentation to the original news propagation graphs. Subsequently, these augmented graphs are encoded using a graph encoder and subsequently reconstructed via a graph decoder. Finally, a composite loss function that encompasses both reconstruction error and contrastive loss is designed. Firstly, it ensures the model can effectively capture the latent features, based on minimizing the discrepancy between reconstructed and original graph representations. Secondly, it aligns the representations of augmented graphs that originate from the same source. Experiments on the real-world dataset validate the effectiveness of our method.

Recently, the autoregressive framework based on large language models (LLMs) has achieved excellent performance in controlling the generated text to adhere to the required style. These methods guide LLMs through prompt learning to generate target text in an autoregressive manner. However, this manner possesses lower controllability and suffers from the challenge of accumulating errors, where early prediction inaccuracies might influence subsequent word generation. Furthermore, existing prompt-based methods overlook specific region editing, resulting in a deficiency of localized control over input text. To overcome these challenges, we propose a novel three-stage prompt-based approach for specific region editing. To alleviate the issue of accumulating errors, we transform the text style transfer task into a text infilling task, guiding the LLMs to modify only a small portion of text within the editing region to achieve style transfer, thus reducing the number of autoregressive iterations. To achieve an effective specific editing region, we adopt both prompt-based and word frequency-based strategies for region selection, subsequently employing a discriminator to validate the efficacy of the selected region. Experiments conducted on several publicly competitive datasets for text style transfer task confirm that our proposed approach achieves state-of-the-art performance. Keywords: text style transfer, natural language generation, large language models

The existing models for multi-modal fake news detection focus mainly on capturing common similar semantics between different modalities to improve detection performance. However, they ignore the extraction of inconsistent features between these modalities. The intuitive cognition way people identify a piece of fake news is generally to discover if there are inconsistent semantics among news content itself and its comments, which could be abstracted as “comparing news image-text consistency - finding valuable comments - reasoning in-/consistency between news and comments”. Inspired by the cognitive process, we propose Human Cognition-based Consistency Inference Networks (HCCIN) to comprehensively explore consistent and inconsistent semantics for multi-modal fake news detection. Specifically, we first design cross-modal alignment layer to learn consistent semantics between textual and visual information within the multi-modal news, and then the comment clue discovery layer is devoted to ascertaining the most-concerned semantics by audiences between comments. Finally, we develop collaborative inference layer to drive news consistent semantics and the most-concerned semantics to reason and discover consistent and inconsistent information between them. Experiments on three public datasets, including Weibo, Twitter, and PHEME, reveal the superiority of our HCCIN.

Existing studies on table-based fact verification generally capture linguistic evidence from claim-table subgraphs or logical evidence from program-table subgraphs independently. However, there is insufficient association interaction between the two types of evidence, which makes it difficult to obtain valuable consistency features between them. In this work, we propose heuristic heterogeneous graph reasoning networks (H2GRN) to capture the shared consistent evidence by strengthening associations between linguistic and logical evidence from two perspectives of graph construction and reasoning mechanism. Specifically, 1) to enhance the close connectivity of the two subgraphs, rather than simply connecting two subgraphs by the nodes with the same content (the constructed graph in this way has severe sparsity), we construct a heuristic heterogeneous graph, which relies on claim semantics as heuristic knowledge to guide the connections of the program-table subgraph, and in turn expands the connectivity of the claim-table subgraph through logical information of programs as heuristic knowledge; and 2) to establish adequate association interaction between linguistic evidence and logical evidence, we design multiview reasoning networks. In detail, we propose local-view multihop knowledge reasoning (MKR) networks to enable the current node to establish association not only with one-hop neighbors, but also with multihop neighbors, to capture context-richer evidence information. We execute MKR on heuristic claim-table and program-table subgraphs to learn context-richer linguistic evidence and logical evidence, respectively. Meanwhile, we develop global-view graph dual-attention networks (DAN) that execute on the entire heuristic heterogeneous graph, reinforcing global-level significant consistency evidence. Finally, the consistency fusion layer is devised to weaken the disagreement between the three types of evidence to assist in capturing consistent shared evidence for verifying claims. Experiments on TABFACT and FEVEROUS demonstrate the effectiveness of H2GRN.

The existing approaches based on different neural networks automatically capture and fuse the multimodal semantics of news, which have achieved great success for fake news detection. However, they still suffer from the limitations of both shallow fusion of multimodal features and less attention to the inconsistency between different modalities. To overcome them, we propose multi-reading habits fusion reasoning networks (MRHFR) for multi-modal fake news detection. In MRHFR, inspired by people's different reading habits for multimodal news, we summarize three basic cognitive reading habits and put forward cognition-aware fusion layer to learn the dependencies between multimodal features of news, so as to deepen their semantic-level integration. To explore the inconsistency of different modalities of news, we develop coherence constraint reasoning layer from two perspectives, which first measures the semantic consistency between the comments and different modal features of the news, and then probes the semantic deviation caused by unimodal features to the multimodal news content through constraint strategy. Experiments on two public datasets not only demonstrate that MRHFR not only achieves the excellent performance but also provides a new paradigm for capturing inconsistencies between multi-modal news.

The existing data-driven approaches typically capture credibility-indicative representations from relevant articles for fake news detection, such as skeptical and conflicting opinions. However, these methods still have several drawbacks: 1) Due to the difficulty of collecting fake news, the capacity of the existing datasets is relatively small; and 2) there is considerable unverified news that lacks conflicting voices in relevant articles, which makes it difficult for the existing methods to identify their credibility. Especially, the differences between true and fake news are not limited to whether there are conflict features in their relevant articles, but also include more extensive hidden differences at the linguistic level, such as the perspectives of emotional expression (like extreme emotion in fake news), writing style (like the shocking title in clickbait), etc., the existing methods are difficult to fully capture these differences. To capture more general and wide-ranging differences between true and fake news, in this paper, directly from the different categories of news itself, we propose a Category-controlled Encoder-Decoder model (CED) to generate examples with category-differentiated features and extend the dataset capacity to achieve data enhancement effect, thus enhancing fake news detection. Specifically, to make the generated examples enrich more news features, we develop news-guided encoder to guide relevant articles to generate news-semantic context representations. To drive the generated examples to contain more category-differentiated features, we devise category-controlled decoder which relies on pattern-shared unit to respectively capture intra-category shared features within true or fake news, and employs restriction unit to force the two types of shared features to be more different for highlighting inter-category differentiated features. The experimental results on three datasets demonstrate the superiority of CED.

The domain of Aspect-based Sentiment Analysis, in which aspects are extracted, their sentiments are analysed and sentiments are evolved over time, is getting much attention with increasing feedback of public and customers on social media. The immense advancements in this field urged the researchers to devise new techniques and approaches, each sermonizing a different research analysis/question, that cope with upcoming issues and complex scenarios of Aspect-based Sentiment Analysis. Therefore, this survey emphasized on the issues and challenges that are related to extraction of different aspects and their relevant sentiments, relational mapping between aspects, interactions, dependencies, and contextual-semantic relationships between different data objects for improved sentiment accuracy, and prediction of sentiment evolution dynamicity. A rigorous overview of the recent progress is summarized based on whether they contributed towards highlighting and mitigating the issue of Aspect Extraction, Aspect Sentiment Analysis or Sentiment Evolution. The reported performance for each scrutinized study of Aspect Extraction and Aspect Sentiment Analysis is also given, showing the quantitative evaluation of the proposed approach. Future research directions are proposed and discussed, by critically analysing the presented recent solutions, that will be helpful for researchers and beneficial for improving sentiment classification at aspect-level.

One of the interesting trending phenomena in sentiment analysis is the prediction of sentiment given by the user towards an aspect term. Till today, a considerable number of researchers have proposed varying methodologies for predicting aspect-based sentiments. But they mostly encapsulate the semantic information by manifesting themselves within a local boundary around each aspect term and overlook capturing the semantic concept that is conveyed within the entire review (global). Therefore, this study proposes a model,IAF-LG, that performs semantic learning at both local and global scales to discover aspect-based sentiments. IAF-LG first encodes the local semantics by fusing contextual-semantic dependencies between tokens and computing relational semantics between inter-aspects. Next, it develops the global semantics by formulating interactions between local semantics and review-based sentiment learning. Lastly, it conjoins the local and global interactive learning to earn credible semantics for predicting the accurate sentiment of aspect terms. Extensive experiments on publicly available datasets demonstrate the significantly improved performance of IAF-LG than competitive baselines.

Due to the high maneuverability and strong adaptability, autonomous unmanned aerial vehicles (UAVs) are of high interest to many civilian and military organizations around the world. Automatic path planning which autonomously finds a good enough path that covers the whole area of interest, is an essential aspect of UAV autonomy. In this study, we focus on the automatic path planning of heterogeneous UAVs with different flight and scan capabilities, and try to present an efficient algorithm to produce appropriate paths for UAVs. First, models of heterogeneous UAVs are built, and the automatic path planning is abstracted as a multi-constraint optimization problem and solved by a linear programming formulation. Then, inspired by the density-based clustering analysis and symbiotic interaction behaviours of organisms, an adaptive clustering-based algorithm with a symbiotic organisms search-based optimization strategy is proposed to efficiently settle the path planning problem and generate feasible paths for heterogeneous UAVs with a view to minimizing the time consumption of the search tasks. Experiments on randomly generated regions are conducted to evaluate the performance of the proposed approach in terms of task completion time, execution time and deviation ratio.

Existing approaches construct appropriate interaction models to explore semantic conflicts between claims and relevant articles, which provides practical solutions for interpretable claim verification. However, these conflicts are not necessarily all about questioning the false part of claims, which makes considerable semantic conflicts difficult to be used as evidence to explain the results of claim verification. In this paper, we propose evidence inference networks (EVIN), which focus on the conflicts questioning the core semantics of claims and serve as evidence for interpretable claim verification. Specifically, EVIN first captures the core semantic segments of claims and the users' principal opinions in relevant articles. Then, it finely-grained identifies the semantic conflicts contained in each relevant article from these opinions. Finally, it constructs coherence modeling to match the conflicts that queries the core semantic fragments of claims as explainable evidence. Experiments on two widely used datasets demonstrate that EVIN not only achieves satisfactory performance but also provides explainable evidence for end-users.

Lianwei Wu, Yuan Rao, Yuqian Lan, Ling Sun, Zhaoyin Qi. 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.

Recently, many methods discover effective evidence from reliable sources by appropriate neural networks for explainable claim verification, which has been widely recognized.However, in these methods, the discovery process of evidence is nontransparent and unexplained.Simultaneously, the discovered evidence only roughly aims at the interpretability of the whole sequence of claims but insufficient to focus on the false parts of claims.In this paper, we propose a Decision Tree-based Co-Attention model (DTCA) to discover evidence for explainable claim verification.Specifically, we first construct Decision Tree-based Evidence model (DTE) to select comments with high credibility as evidence in a transparent and interpretable way.Then we design Co-attention Self-attention networks (CaSa) to make the selected evidence interact with claims, which is for 1) training DTE to determine the optimal decision thresholds and obtain more powerful evidence; and 2) utilizing the evidence to find the false parts in the claim.Experiments on two public datasets, RumourEval and PHEME, demonstrate that DTCA not only provides explanations for the results of claim verification but also achieves the state-of-the-art performance, boosting the F1-score by 3.11%, 2.41%, respectively.

The majority of existing methods for fake news detection universally focus on learning and fusing various features for detection. However, the learning of various features is independent, which leads to a lack of cross-interaction fusion between features on social media, especially between posts and comments. Generally, in fake news, there are emotional associations and semantic conflicts between posts and comments. How to represent and fuse the cross-interaction between both is a key challenge. In this paper, we propose Adaptive Interaction Fusion Networks (AIFN) to fulfill cross-interaction fusion among features for fake news detection. In AIFN, to discover semantic conflicts, we design gated adaptive interaction networks (GAIN) to capture adaptively similar semantics and conflicting semantics between posts and comments. To establish feature associations, we devise semantic-level fusion self-attention networks (SFSN) to enhance semantic correlations and fusion among features. Extensive experiments on two real-world datasets, i.e., RumourEval and PHEME, demonstrate that AIFN achieves the state-of-the-art performance and boosts accuracy by more than 2.05% and 1.90%, respectively.

Exploring evidence from relevant articles to confirm the veracity of claims is a trend towards explainable claim verification. However, most strategies capture the top-k check-worthy articles or salient words as evidence, but this evidence is difficult to focus on the questionable parts of unverified claims. Besides, they utilize relevant articles indiscriminately, ignoring the source credibility of these articles, which may cause quiet a few unreliable articles to interfere with the assessment results. In this paper, we propose Evidence-aware Hierarchical Interactive Attention Networks (EHIAN) by considering the capture of evidence fragments and the fusion of source credibility to explore more credible evidence semantics discussing the questionable parts of claims for explainable claim verification. EHIAN first designs internal interaction layer (IIL) to strengthen deep interaction and matching between claims and relevant articles for obtaining key evidence fragments, and then proposes global inference layer (GIL) that fuses source features of articles and interacts globally with the average semantics of all articles and finally earns the more credible evidence semantics discussing the questionable parts of claims. Experiments on two datasets demonstrate that EHIAN not only achieves the state-of-the-art performance but also secures effective evidence to explain the results.

Lianwei Wu, Yuan Rao, Haolin Jin, Ambreen Nazir, Ling Sun. 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.