Existing neural-training-free AI-Generated Text (AIGT) detectors usually aggregate token-level probability signals with uniform weights. However, this ignores an important linguistic regularity: human-written and AI-generated texts exhibit different Part-Of-Speech (POS) n-gram patterns, making certain tokens more informative than others. To address the limitations, we propose SWLD-Joint (Syntax-pattern Weighted Log-likelihood Difference Joint Detection), a calibration-based neural-training-free detector without gradient-based optimization or neural classifier fine-tuning. SWLD-Joint first estimates POS n-gram template weights from labeled calibration data and then combines them with a POS-trajectory calibration signal to focus on syntactically informative positions. Experiments on DetectRL, RAID, and HC3_Ext show robust gains under domain shift, generator mismatch, and adversarial perturbations. Ablations, parameter analysis, case studies, score distributions, and ROC curves verify the effectiveness of syntax-pattern weighting. Code and data will be released upon acceptance.

Rule-based methods for temporal knowledge graph (TKG) forecasting are valued for the interpretability of their explicit logic chains. However, they suffer from a severe cold-start coverage gap: exact-match closed walks terminate at sparse entities, leaving many queries without any rule evidence. Existing similarity-based remedies are unreliable because their similarity is derived from static structural co-occurrence or shallow embeddings with no temporal calibration and hand-tuned thresholds. To address these limitations, we propose SimRule, a collaborative recommendation framework for TKG rule mining. SimRule first mines both exact rules and collaborative-surrogate rules through a temporally calibrated dual-view entity similarity index. An adaptive dual-branch fusion then ranks candidate objects from their aggregated evidence, and each forecast remains fully auditable as an explicit rule chain with quantified time gaps. On ICEWS14, ICEWS18, and ICEWS05-15, SimRule achieves state-of-the-art MRR and Hits@K (K ≥ 3) among rule-based methods.

Hallucinations undermine the reliability of Large Language Models (LLMs) in real-world settings. Current contrastive decoding methods mitigate hallucinations by generating multiple responses based on an amateur model and an expert model or using the logit distribution across all layers, leading to considerable redundancy and computational inefficiency during inference. To address these limitations, we propose SHICD (Self-Hallucination-Inducing Contrastive Decoding), a novel training-free decoding framework that requires a single response from an individual model and leverages information from a subset of all layers, without any external intervention on activations. Specifically, SHICD first identifies the inducing layer of the LLM when an initial semantic representation emerges in the model. The hallucination direction is then extracted to induce hallucinations by itself while preserving the initial semantic information encoded before the inducing layer. Finally, we employ an adaptive contrastive decoding mechanism to determine when to apply contrastive decoding during generation. Extensive experiments across multiple LLMs and benchmark datasets demonstrate that SHICD significantly improves factual accuracy over baseline methods while achieving lower inference latency.

The rise of multimodal content on social platforms has led to the rapid spread of complex and persuasive false narratives, combining of text and images. Traditional rumor detection models attempt to identify such content by relying on textual cues or employing shallow multimodal fusion techniques. However, these methods often assume a simplistic one-to-one alignment between modalities, overlooking the richer hierarchical relationships across modalities, failing to capture the layered structure of meaning. In this paper, we present RumorCone, a novel method that employs hyperbolic geometry in order to preserve hierarchical, non-linear relationships, rather than representing them at a flat semantic level. First, RumorCone decomposes image and text content into three levels: base, mid, and high-level abstractions, and embeds them in hyperbolic space to model their tree-like semantic structure. Second, a dynamic hyperbolic multimodal attention mechanism aligns features across modalities and levels, and a flexible fusion strategy adjusts the contribution of each modality based on alignment quality. Our experiments indicate the importance of hierarchical semantic modeling for robust and interpretable multimodal rumor detection.

Patent classification is a fundamental step in the patent examination process, directly impacting the efficiency and quality of substantive review. Existing methods mostly focus on general texts like titles and abstracts, thus ignoring the key technical content claims and the corresponding citation relationships. Meanwhile, these approaches treat labels as independent targets, failing to exploit the semantic and structural information within the label taxonomy. To address these problems, we propose a Claim Structure based Patent Classification model with Label Awareness (CSPC-LA). The method first utilizes the citation relationship of patent claim texts to construct the citation graph and the co-reference graph. Then structural graph learning is used on both graphs to mine the internal logic of patent claims. Finally, we optimize the tree hierarchy of IPC labels and employ tree propagation learning to enhance the patent representation. Extensive experiments on the latest patent classification dataset from USPTO demonstrate that the proposed method is more effective than the state-of-the-art baselines.

Clickbait posts tend to spread inaccurate or misleading information to manipulate people’s attention and emotions, which greatly harms the credibility of social media. Existing clickbait detection models rely on analyzing the objective semantics in posts or correlating posts with article content only. However, these models fail to identify and exploit the manipulation intention of clickbait from a user’s subjective perspective, leading to limited capability to explore comprehensive clues of clickbait. To address such a issue, we propose a multiview clickbait detection model, named MCDM, to model subjective and objective preferences simultaneously. MCDM introduces two novel complementary modules for modeling subjective feeling and objective content relevance, respectively. The subjective feeling module adopts a user-centric approach to capture subjective features of posts, such as language patterns and emotional inclinations. The objective module explores news elements from posts and models article content correlations to capture objective clues for clickbait detection. Extensive experimental results on two real-world datasets show that our proposed MCDM outperforms state-of-the-art approaches for clickbait detection, verifying the effectiveness of integrating subjective and objective preferences for detecting clickbait.

Abstractive related work generation has attracted increasing attention in generating coherent related work that helps readers grasp the current research. However, most existing models ignore the inherent causality during related work generation, leading to spurious correlations which downgrade the models’ generation quality and generalizability. In this study, we argue that causal intervention can address such limitations and improve the quality and coherence of generated related work. To this end, we propose a novel Causal Intervention Module for Related Work Generation (CaM) to effectively capture causalities in the generation process. Specifically, we first model the relations among the sentence order, document (reference) correlations, and transitional content in related work generation using a causal graph. Then, to implement causal interventions and mitigate the negative impact of spurious correlations, we use do-calculus to derive ordinary conditional probabilities and identify causal effects through CaM. Finally, we subtly fuse CaM with Transformer to obtain an end-to-end related work generation framework. Extensive experiments on two real-world datasets show that CaM can effectively promote the model to learn causal relations and thus produce related work of higher quality and coherence.

Joint intent detection and slot filling has recently achieved tremendous success in advancing the performance of utterance understanding. However, many joint models still suffer from the robustness problem, especially on noisy inputs or rare/unseen events. To address this issue, we propose a Joint Adversarial Training (JAT) model to improve the robustness of joint intent detection and slot filling, which consists of two parts: (1) automatically generating joint adversarial examples to attack the joint model, and (2) training the model to defend against the joint adversarial examples so as to robustify the model on small perturbations. As the generated joint adversarial examples have different impacts on the intent detection and slot filling loss, we further propose a Balanced Joint Adversarial Training (BJAT) model that applies a balance factor as a regularization term to the final loss function, which yields a stable training procedure. Extensive experiments and analyses on the lightweight models show that our proposed methods achieve significantly higher scores and substantially improve the robustness of both intent detection and slot filling. In addition, the combination of our BJAT with BERT-large achieves state-of-the-art results on two datasets.