Emotion-Cause Pair Extraction in Conversations (ECPEC) aims to identify the set of causal relations between emotion utterances and their triggering causes within a dialogue. Most existing approaches formulate ECPEC as independent pairwise classification, overlooking the distinct semantics of emotion diffusion and cause explanation, and failing to capture globally consistent many-to-many conversational causality. To address these limitations, we revisit ECPEC from a semantic perspective and seek to disentangle emotion-oriented semantics from cause-oriented semantics, mapping them into two complementary representation spaces to better capture their distinct conversational roles. Building on this semantic decoupling, we naturally formulate ECPEC as a global alignment problem between the emotion-side and cause-side representations, and employ optimal transport to enable many-to-many and globally consistent emotion-cause matching. Based on this perspective, we propose a unified framework SCALE that instantiates the above semantic decoupling and alignment principle within a shared conversational structure. Extensive experiments on several benchmark datasets demonstrate that SCALE consistently achieves state-of-the-art performance.
论文检索
输入标题、作者或关键词，从 100,903 篇学术成果中精准定位
Probing Multimodal Large Language Models on Cognitive Biases in Chinese Short-Video Misinformation
PDF ↗Short-video platforms have become major channels for misinformation, where deceptive claims frequently leverage visual experiments and social cues. While Multimodal Large Language Models (MLLMs) have demonstrated impressive reasoning capabilities, their robustness against misinformation entangled with cognitive biases remains under-explored. In this paper, we introduce a comprehensive evaluation framework using a high-quality, manually annotated dataset of 200 short videos spanning four health domains. This dataset provides fine-grained annotations for three deceptive patterns—experimental errors, logical fallacies, and fabricated claims—each verified by evidence such as national standards and academic literature. We evaluate eight frontier MLLMs across five modality settings. Experimental results demonstrate that Gemini-2.5-Pro achieves the highest performance in the multimodal setting with a belief score of 71.5/100, while o3 performs the worst at 35.2. Furthermore, we investigate social cues that induce false beliefs in videos and find that models are susceptible to biases like authoritative channel IDs.
Going beyond the prediction of numerical scores, recent research in automated essay scoring has increasingly emphasized the generation of high-quality feedback that provides justification and actionable guidance. To mitigate the high cost of expert annotation, prior work has commonly relied on LLM-generated feedback to train essay assessment models. However, such feedback is often incorporated without explicit quality validation, resulting in the propagation of noise in downstream applications. To address this limitation, we propose FeedEval, an LLM-based framework for evaluating LLM-generated essay feedback along three pedagogically grounded dimensions: specificity, helpfulness, and validity. FeedEval employs dimension-specialized LLM evaluators trained on datasets curated in this study to assess multiple feedback candidates and select high-quality feedback for downstream use. Experiments on the ASAP++ benchmark show that FeedEval closely aligns with human expert judgments and that essay scoring models trained with FeedEval-filtered high-quality feedback achieve superior scoring performance. Furthermore, revision experiments using small LLMs show that the high-quality feedback identified by FeedEval leads to more effective essay revisions. We release our code and curated datasets at: https://github.com/BBeeChu/FeedEval.git.
Existing video understanding benchmarks mainly emphasize general visual recognition and reasoning, but do not adequately capture the pedagogical logic embedded in instructional videos. To address this gap, we present PedagogyBench, a multimodal benchmark for instructional video understanding grounded in pedagogical cognition. We introduce a pedagogy-driven segmentation strategy and a dual-stream semantic injection pipeline that combines machine pre-annotation with expert refinement, enabling the construction of a dataset organized around a cognitive pyramid with four levels and 20 fine-grained tasks. We further propose the Cognitive Fidelity Score (CFS) to measure the balance of model performance across pedagogical cognitive dimensions. Experiments on 12 multimodal large language models reveal a clear generative gap, where models perform relatively well on discriminative tasks but degrade on higher-order pedagogical diagnosis, often relying on parametric memory rather than grounded visual perception. Project resources are available at https://github.com/Shallcom/PedagogyBench.
Advances in Multimodal Large Language Models (MLLMs) intensify concerns about data safety, making Machine Unlearning (MU), the selective removal of harmful/private information, a critical necessity. However, existing MU benchmarks for MLLMs are limited by a lack of image diversity, coarse-grained unlearning target, and insufficient evaluation scenarios, which fail to capture the complexity of real-world applications. To facilitate the development of MLLMs unlearning and alleviate the aforementioned limitations, we introduce OFFSIDE, a novel benchmark for evaluating misinformation unlearning in MLLMs. This manually curated dataset contains 15.68K records for 80 players, providing a comprehensive framework with four test sets to assess forgetting efficacy, generalization, utility, and robustness. OFFSIDE supports advanced unlearning targets, such as fine-grained unlearning and visual rumor removal. Our extensive evaluation of multiple baselines not only extends key findings from LLM MU to MLLM MU: (1) unlearned rumors can be easily recovered through relearning and (2) all methods are vulnerable to prompt attacks, but also introduces novel insights in the context of MLLM: (1) unimodal methods fail to handle multimodal rumors, (2) unlearning efficacy is primarily driven by catastrophic forgetting statistically, and (3) all methods struggle with visual rumors (rumors embedded in images). These results expose significant vulnerabilities in current approaches, highlighting the need for more robust multimodal unlearning solutions.
Retrieval shapes how language models access and cite knowledge in retrieval-augmented generation (RAG). In historical research, the goal is often to locate the exact record for a specific regnal month, where temporal alignment matters as much as topical relevance. This is especially challenging for Classical Chinese annals: time is encoded in terse, implicit, non-Gregorian reign phrases that are context-dependent, so semantically plausible evidence can still be temporally invalid. We introduce **ChunQiuTR**, a time-keyed retrieval benchmark built from the **Spring and Autumn Annals** and its exegetical tradition. It organizes records by month-level reign keys and includes chrono-near confounders that mimic real retrieval failures. We propose **CTD** (Calendrical Temporal Dual-encoder), a time-aware dual-encoder combining Fourier-based absolute context with relative offset biasing. Experiments show consistent gains over semantic dual-encoder baselines under time-keyed evaluation. We will release ChunQiuTR and code after the anonymity period.
Redefining Machine Simultaneous Interpretation: From Incremental Translation to Human-Like Strategies
PDF ↗Simultaneous Machine Translation (SiMT) requires high-quality translations under strict real-time constraints, which traditional policies with only READ/WRITE actions cannot fully address. We extend the action space of SiMT with four adaptive actions: **Sentence_Cut**, **Drop**, **Partial_Summarization** and **Pronominalization**, which enable real-time restructuring, omission, and simplification while preserving semantic fidelity. We adapt these actions in a large language model (LLM) framework and construct training references through action-aware prompting. To evaluate both quality and word-level monotonicity, we further develop a latency-aware TTS pipeline that maps textual outputs to speech with realistic timing. Experiments on the ACL60/60 English-Chinese, English-German and English-Japanese benchmarks show that our framework consistently improves semantic metrics and achieves lower delay compared to reference translations and salami-based baselines. Notably, combining **Drop** and **Sentence_Cut** leads to consistent improvements in the balance between fluency and latency. These results demonstrate that enriching the action space of LLM-based SiMT provides a promising direction for bridging the gap between human and machine interpretation.
Balancing Classification and Calibration Performance in Decision-Making LLMs via Calibration Aware Reinforcement Learning
PDF ↗Large language models (LLMs) are increasingly deployed in decision-making tasks, where not only accuracy but also reliable confidence estimates are essential. Well-calibrated confidence enables downstream systems to decide when to trust a model and when to defer to fallback mechanisms. In this work, we conduct a systematic study of calibration in two widely used fine-tuning paradigms: supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). We show that while RLVR improves task performance, it produces extremely overconfident models, whereas SFT yields substantially better calibration, even under distribution shift, though with smaller performance gains. Through targeted experiments, we diagnose RLVR’s failure, showing that decision tokens act as extraction steps of the decision in reasoning traces and do not carry confidence information, which prevents reinforcement learning from surfacing calibrated alternatives. Based on this insight, we propose a calibration-aware reinforcement learning formulation that directly adjusts decision-token probabilities. Our method preserves RLVR’s accuracy level while mitigating overconfidence, reducing ECE scores up to 9 points.
Watch Out Your Industrial Copilots: Stealthy Backdoor Attack Against LLM-Based PLC Code Generation
PDF ↗Recently, there is an emerging trend of using Large Language Models (LLMs) to generate Programmable Logic Controller (PLC) code automatically, resulting in commercialized products such as Siemens Industrial Copilots. While such LLM-driven products have the potential to transform the way control engineers program, they may also introduce a new attack surface. In this work, we introduce STBack, the first stealthy backdoor attack framework targeting LLM-based PLC code generation. STBack first incorporates six malicious logic injection patterns specifically designed for PLCs to generate the poisoned code samples, along with a three-stage automated pipeline to refine stealthiness. Then, it injects the backdoor by finetuning an LLM using the prompts with a semantic-integrated trigger and the corresponding malicious PLC code sample pairs. The compromised LLM will generate malicious PLC code when the trigger is identified in the prompts.We evaluate STBack on multiple LLMs, which achieves 82.92% average attack success rate while remaining stealthy, i.e., maintaining over 95% semantic similarity with benign code and bypassing quality validation, making the injected backdoor extremely challenging to detect. We also show that existing defenses are ineffective against our benign-looking trigger mechanism. This work reveals a novel and critical security threat for industrial copilots, calling for more cautious use and dedicated defenses.
RADS: Reinforcement Learning-Based Sample Selection Improves Transfer Learning in Low-resource and Imbalanced Clinical Settings
PDF ↗A common strategy in transfer learning is few shot fine-tuning, but its success is highly dependent on the quality of samples selected as training examples. Active learning methods such as uncertainty sampling and diversity sampling can select useful samples. However, under extremely low-resource and class-imbalanced conditions, they often favor outliers rather than truly informative samples, resulting in degraded performance. In this paper, we introduce RADS (Reinforcement Domain Adaptive Sampling), a robust sample selection strategy using reinforcement learning (RL) to identify the most informative samples. Experimental evaluations on several real world clinical datasets show our sample selection strategy enhances model transferability while maintaining robust performance under extreme class imbalance compared to traditional methods. Our code is open-sourced on GitHub.
Causal-Audit: Explicit and Auditable Graph-based Reasoning via Target-Aware Causal Chain Construction
PDF ↗Causal and intervention-based question answering is fundamental to advancing large language models (LLMs) toward reasoning beyond surface-level correlations and understanding underlying causal mechanisms. However, existing LLM-based methods often rely on implicit language-level reasoning, resulting in opaque causal assumptions, unverifiable reasoning paths, and fragile predictions under complex interventions, particularly in context-free settings. In this paper, we propose an explicit and auditable causal reasoning framework for context-free intervention-based question answering. Our method formulates causal inference as structured reasoning over an explicit causal graph through four modular stages, rather than implicit end-to-end prediction. A key innovation is a target-aware causal graph construction strategy that treats the target variable as a core constraint during graph expansion, effectively suppressing irrelevant variables, spurious causal relations, and reasoning noise. We further introduce a path-level causal evidence aggregation mechanism that combines multiple causal paths while modeling both reinforcing and counteracting effects, enabling robust decision-making beyond single-chain reasoning. Extensive experiments on three benchmarks demonstrate that our framework consistently outperforms existing LLM-based methods while providing interpretable and auditable causal reasoning traces.
SplitThenMerge: Token-Level Skill-Compositional Sparse Mixture-of-Experts for Complex Domain-Specific Tasks
PDF ↗Large language models have demonstrated strong performance on general-purpose tasks but often fail to satisfy the accuracy requirements of knowledge-intensive domains such as law, medicine, and finance. Complex domain-specific generation is inherently compositional, involving multiple atomic skills such as reasoning, knowledge grounding, and numerical computation that are frequently interleaved at the token level. Existing domain adaptation methods typically train these heterogeneous skills jointly within a single objective, which makes it difficult for models to reliably coordinate multiple skills when solving complex tasks. In this work, we explicitly incorporate atomic skills into domain-specific model training and propose SplitThenMerge, a framework that decomposes domain competence into atomic skills, trains them independently, and composes them dynamically during generation. SplitThenMerge adopts a token-level sparse Mixture-of-Experts architecture to enable fine-grained skill routing and coordination while implementing each skill as a lightweight LoRA expert to achieve parameter-efficient specialization. Experimental results demonstrate that our method consistently achieves superior performance in both legal and medical domains under the same training parameter budget.
Simulating dementia patients with large language models (LLMs) is challenging due to the need to jointly model cognitive impairment, emotional dynamics, and nonverbal behaviors over long conversations. We present DemMA, an expert-guided dementia dialogue agent for high-fidelity multi-turn patient simulation. DemMA constructs clinically grounded dementia personas by integrating pathology information, personality traits, and subtype-specific memory-status personas informed by clinical experts. To move beyond text-only simulation, DemMA explicitly models nonverbal behaviors, including motion, facial expressions, and vocal cues. We further introduce a Chain-of-Thought distillation framework that trains a single LLM to jointly generate reasoning traces, patient utterances, and aligned behavioral actions within one forward pass, enabling efficient deployment without multi-agent inference.
LEAF: Towards Lightweight Explainable Hateful Video Detection via Self-Grounding CoT Guided Stage-Wise Distillation
PDF ↗The rapid spread of hateful videos online has sparked growing social concerns, driving research efforts to detect and limit their dissemination. However, existing methods rely on opaque models that offer no insight into their decisions, eroding trust in detection systems. Large Multimodal Models (LMMs) provide a compelling alternative, thanks to their ability to generate free-text explanations for multimodal content. Yet, their high computational demands and pronounced bias toward benign predictions limit their practicality. We introduce LEAF, the first Lightweight, Explainable hAteful video detection Framework. At its core, LEAF distills the "explainability" from LMMs into efficient Smaller Multimodal Models (SMMs) through a controlled, de-biasing process, enabling lightweight yet interpretable Hateful Video Detection (HVD). We achieve this with a novel Self-Grounding Chain-of-Thought mechanism that guides LMMs to generate high-quality, unbiased explanatory supervision signals for videos. These signals then progressively train the SMM via a new Stage-Wise Distillation paradigm, resulting in faithful, human-readable natural language explanations for HVD. Extensive experiments on three video benchmarks demonstrate that LEAF not only outperforms prior methods in detection accuracy but also provides strong explainability — all with a lightweight design.
Reinforcement Learning with Verifiable Rewards (RLVR) is an essential paradigm that enhances the reasoning capabilities of Large Language Models (LLMs). However, existing methods typically rely on static policy optimization schemes that misalign with the model’s evolving reasoning capabilities. To address this issue, we propose Adaptive Power-Mean Policy Optimization (APMPO), which comprises two main innovations: Power-Mean Policy Optimization (PMPO) and Feedback-Adaptive Clipping (FAC). Specifically, PMPO introduces a generalized power-mean objective. This enables the model to adaptively transition from the signal-amplifying behavior of the arithmetic mean to the consistency-enforcing behavior of the geometric mean. FAC adaptively adjusts clipping bounds based on real-time reward statistics to overcome the limitations of static mechanisms. Capitalizing on these innovations, APMPO improves learning dynamics and reasoning performance. Extensive experiments on nine datasets across three reasoning tasks showcase the superiority of APMPO over state-of-the-art RLVR-based baselines. For instance, APMPO boosts the average Pass@1 score on mathematical reasoning benchmarks by 3.0 points compared to GRPO when using Qwen2.5-3B-Instruct.
Reward-guided search methods have demonstrated strong potential in enhancing tool-using agents by effectively guiding sampling and exploration over complex action spaces. As a core design, those search methods utilize process reward models (PRMs) to provide step-level rewards, enabling more fine-grained monitoring. However, there is a lack of systematic and reliable evaluation benchmarks for PRMs in tool-use settings. In this paper, we introduce ToolPRMBench, a large-scale benchmark specifically designed to evaluate PRMs for tool-using agents. ToolPRMBench is built on top of several representative tool-use benchmarks and converts agent trajectories into step-level test cases. Each case contains the interaction history, a correct action, a plausible but incorrect alternative, and relevant tool metadata. We respectively utilize offline sampling to isolate local single-step errors and online sampling to capture realistic multi-step failures from full agent rollouts. A multi-LLM verification pipeline is proposed to reduce label noise and ensure data quality. We conduct extensive experiments across large language models, general PRMs, and tool-specialized PRMs on ToolPRMBench. The results reveal clear differences in PRM effectiveness and highlight the potential of specialized PRMs for tool-using. Our code and dataset are available at: https://github.com/David-Li0406/ToolPRMBench[More resources on LLM-as-a-judge are on the website: <https://llm-as-a-judge.github.io>].
Preference optimization is fundamental for aligning large language models. While existing methods use sample weighting, they typically rely on static functions of instantaneous model states and ignore temporal learning dynamics. We contend that a sample’s value evolves throughout training, characterized by patterns such as stable convergence or noisy oscillation. We propose MetaPO, a framework that meta-learns adaptive weights using three temporal features: reward margin evolution, learning volatility, and reference deviation. Through bilevel optimization on validation data, MetaPO automatically discovers weighting strategies tailored to specific datasets. Experiments on models ranging from 7B to 70B parameters demonstrate statistically significant improvements over strong baselines, achieving gains of up to 2.4 points on AlpacaEval 2.0 and Arena-Hard. Interpretability analysis confirms that temporal features drive over 70% of the weighting decisions and that the learned weights correlate strongly with sample quality.
Defending Large Language Models (LLMs) against backdoor attacks has long been trapped in a "cat-and-mouse" dilemma, where defenders passively react to ever-shifting attack strategies. To break this cycle, we posit that proactive immunization is inherently superior to reactive sanitization. In this study, we propose Poison-to-Poison (P2P), a general and effective defense algorithm that introduces a paradigm shift. Instead of waiting to detect malicious samples, P2P strategically implants benign triggers to reshape the model’s decision boundary, redirecting latent feature activation from malicious trajectories to a safe, controllable output space. This enforces the model to associate trigger-induced representations with safe outputs, thereby overriding the effects of original malicious triggers. Thanks to this robust and generalizable trigger-based fine-tuning, P2P is effective across task settings and attack types. Theoretically and empirically, we show that P2P can neutralize malicious backdoors while preserving task performance. We conduct extensive experiments on classification, mathematical reasoning, and summary generation tasks, involving multiple state-of-the-art LLMs. The results demonstrate that our P2P algorithm significantly reduces the attack success rate compared with baseline models. We hope that P2P can serve as a practical guideline for defending against backdoor attacks in the Model as a Service (MaaS) scenario, where benign prompts are embedded within the system to regulate model behavior.
A Herd of Language Models Makes a Better Zero-shot Annotator for Clinical Named Entity Recognition
PDF ↗Clinical named entity recognition (NER) remains difficult to scale due to the high cost of manual annotation. Although large language models (LLMs) enable zero-shot annotation, their performance on clinical NER is still limited. To this end, we improve the annotation quality by aggregating annotations from *a herd of diverse LLMs*, including general-purpose, medically adapted, and NER-specialized models. A key challenge in this multi-LLM setting is effectively leveraging entities extracted by only a minority of models: although they account for a substantial portion of true positives, they are heavily intermixed with noise. To address this, we introduce **MARY**, a label-modeling method for **M**ulti-LLM **A**nnotation using **R**epresentation learning to capture contextual similarit**Y**. During aggregation, MARY selectively incorporates minority-extracted entities whose contexts are similar to those of majority-extracted entities, yielding more reliable and comprehensive annotations. Experimental results show that MARY improves the average F1 score by 8.6% over vanilla zero-shot baselines while reducing annotation costs.
A major challenge for the operation of large language models (LLMs) is how to predict whether a specific LLM will produce sufficiently high-quality output for a given query. Existing approaches rely on external classifiers, most commonly BERT-based models, which suffer from limited context windows, constrained representational capacity, and additional computational overhead. We propose IntroLM, a method that enables causal language models to predict their own output quality during the prefilling phase without affecting generation using [CPX] tokens. By introducing token-conditional LoRA that activates only for the introspective [CPX] token, the model learns to predict the output quality for a given query while preserving the original backbone behavior and avoiding external evaluators. On question-answering benchmarks, IntroLM applied to Qwen3-8B achieves a ROC–AUC of 90% for success prediction, outperforming a DeBERTa-v3-Large classifier by 14%. When integrated into multi-model routing systems, IntroLM achieves superior cost–performance trade-offs, reducing end-to-end latency by up to 33% and large-model usage by up to 50% at matched reliability. Our code is available at https://github.com/hhosseini1377/LLM_routing.