In multi-hop reasoning, multi-round retrieval-augmented generation (RAG) methods typically rely on LLM-generated content as the retrieval query. However, these approaches are inherently vulnerable to knowledge overshadowing—a phenomenon where critical information is overshadowed during generation. As a result, the LLM-generated content may be incomplete or inaccurate, leading to irrelevant retrieval and causing error accumulation during the iteration process. To address this challenge, we propose ActiShade, which detects and activates overshadowed knowledge to guide large language models(LLMs) in multi-hop reasoning. Specifically, ActiShade iteratively detects the overshadowed keyphrase in the given query, retrieves documents relevant to both the query and the overshadowed keyphrase, and generates a new query based on the retrieved documents to guide the next-round iteration. By supplementing the overshadowed knowledge during the formulation of next-round queries while minimizing the introduction of irrelevant noise, ActiShade reduces the error accumulation caused by knowledge overshadowing. Extensive experiments show that ActiShade outperforms existing methods across multiple datasets and LLMs.

Knowledge graphs (KGs) can provide structured knowledge to assist large language models (LLMs) in interpretable reasoning. Knowledge graph question answering (KGQA) is a typical benchmark to evaluate KG-enhanced LLM methods. Previous methods of KG-enhanced LLMs for KGQA mainly include: 1) origin question-oriented methods, which perform KG retrieval based solely on the original question without explicitly analyzing multi-step reasoning logic; and 2) stepwise reasoning-oriented methods, which alternate between LLM generating the next reasoning step and targeted KG retrieval but lack systematic planning, leading to poor controllability. To tackle these limitations, we propose KELGoP, a framework of KG-enhanced LLM based on global planning. We propose fine-grained question categorization based on reasoning patterns and corresponding category-driven question decomposition for complex questions, enabling more controllable reasoning and atomic KG retrieval targeted to sub-questions. Furthermore, we propose an adaptive strategy that allows adjusting the reasoning pattern based on the performance of question answering, making the reasoning more flexible and robust. Finally, we introduce several efficient atomic KG retrieval strategies that operate on KG subgraphs to assist the LLM in answering atomic-level questions. A series of experiments on KGQA datasets demonstrate that our proposed framework achieves superior performance compared to existing baselines.

Large language models (LLMs) have demonstrated significant advancements in code generation, yet they still face challenges when tackling tasks that extend beyond their basic capabilities. Recently, the concept of self-debugging has been proposed as a way to enhance code generation performance by leveraging execution feedback from tests. However, the availability of high-quality tests in real-world scenarios is often limited. In this context, self-debugging with self-generated tests emerges as a promising solution, though its limitations and practical potential have not been fully explored. To address this gap, we investigate the efficacy of self-debugging in code generation tasks. We propose and analyze two distinct paradigms for the self-debugging process: post-execution and in-execution self-debugging. Our findings reveal that post-execution self-debugging struggles with the test bias introduced by self-generated tests, which can lead to misleading feedback. In contrast, in-execution self-debugging enables LLMs to mitigate this bias and leverage intermediate states during program execution. By focusing on runtime information rather than relying solely on potentially flawed self-generated tests, this approach demonstrates significant promise for improving the robustness and accuracy of LLMs in code generation tasks.

Yuhang Tian, Dandan Song, Zhijing Wu, Pan Yang, Changzhi Zhou, Jun Yang, Hao Wang, Huipeng Ma, Chenhao Li, Luan Zhang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Large language models (LLMs) have shown exceptional performance across various domains. However, LLMs are prone to hallucinate facts and generate non-factual responses, which can undermine their reliability in real-world applications. Current hallucination detection methods suffer from external resource demands, substantial time overhead, difficulty overcoming LLMs' intrinsic limitation, and insufficient modeling. In this paper, we propose MHAD, a novel internal-representation-based hallucination detection method. MHAD utilizes linear probing to select neurons and layers within LLMs. The selected neurons and layers are demonstrated with significant awareness of hallucinations at the initial and final generation steps. By concatenating the outputs from these selected neurons of selected layers at the initial and final generation steps, a hallucination awareness vector is formed, enabling precise hallucination detection via an MLP. Additionally, we introduce SOQHD, a novel benchmark for evaluating hallucination detection in Open-Domain QA (ODQA). Extensive experiments show that MHAD outperforms existing hallucination detection methods across multiple LLMs, demonstrating superior effectiveness.

Incorporating explicit personas into dialogue models is critical for generating responses that fulfill specific user needs and preferences, creating a more personalized and engaging interaction. Early works on persona-based dialogue generation directly concatenate the persona descriptions and dialogue history into relatively small pre-trained language models (PLMs) for response generation, which leads to uninformative and inferior results due to the sparse persona information and the limited model generation capabilities. Recently, large language models (LLMs) have shown their surprising capabilities in language generation. Prompting the LLMs with the persona descriptions for role-playing dialogue generation has also achieved promising results. However, deploying LLMs is challenging for practical applications due to their large scale, spurring efforts to distill the generation capabilities into more concise and compact models through teacher-student learning. In this article, we propose an efficient compact K nowledge-grounded P ersona-based D ialogue model enhanced by LLM D istillation (KPDD). Specifically, first, we propose to enrich the annotated persona descriptions by integrating external knowledge graphs (KGs) with a mixed encoding network, coupled with a mixture of experts (MoE) module for both informative and diverse response generation. The mixed encoding network contains multiple layers of modality interaction operations, enabling information from both modalities propagates to the other. Second, to fully exploit the generation capabilities of LLMs, we turn to the distillation technique to improve the generation capabilities of our model, facilitated by a natural language inference (NLI)-based filtering mechanism to extract high-quality information from LLMs. In addition, we employ a curriculum learning strategy to train our model on the high-quality filtered distilled data and progressively on the relatively noisy original data, enhancing its adaptability and performance. Extensive experiments show that KPDD outperforms state-of-the-art baselines in terms of both automatic and human evaluation.

Jing Xu, Dandan Song, Siu Hui, Zhijing Wu, Meihuizi Jia, Hao Wang, Yanru Zhou, Changzhi Zhou, Ziyi Yang. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

The Dialogue-level Aspect-based Sentiment Quadruple analysis (DiaASQ) task has recently received attention in the Aspect-Based Sentiment Analysis (ABSA) field. It aims to extract(target, aspect, opinion, sentiment) quadruples from multi-turn and multi-party dialogues. Compared to previous ABSA tasks focusing on text such as sentences, the DiaASQ task involves more complex contextual information and corresponding relations between terms, as well as longer sequences. These characteristics challenge existing methods that struggle to model explicit span-level interactions or have high computational costs. In this paper, we propose a span-pair interaction and tagging method to solve these issues, which includes a novel Span-pair Tagging Scheme (STS) and a simple and efficient Multi-level Representation Model (MRM). STS simplifies the DiaASQ task to a span-pair tagging task and explicitly captures complete span-level semantics by tagging span pairs. MRM efficiently models the dialogue structure information and span-level interactions by constructing multi-level contextual representation. Besides, we train a span ranker to improve the running efficiency of MRM. Extensive experiments on multilingual datasets demonstrate that our method outperforms existing state-of-the-art methods.

The temporal entity alignment task aims to discover entities with the same meaning but belonging to different temporal knowledge graphs (KGs). Most existing entity alignment studies mainly focus on static entity alignment, while temporal entity alignment has not received enough attention. However, entity alignment containing temporal information is more in line with real-world application scenarios, and applying static entity alignment models directly to temporal KGs usually does not achieve satisfactory performance because many events (entities) in the knowledge graph will change with time. Therefore, we propose an adaptive graph network (AGN) for entity alignment between temporal KGs. Specifically, we use a time-aware graph attention network model as an encoder to aggregate the features and temporal relationships of neighboring nodes. To adapt to various temporal knowledge graphs, we design a training scheme with adaptive relative error loss minimization , which aims to provide relative positions of entities in vector space for model optimization. Furthermore, we propose an adaptive fine-tuning distance algorithm based on supervised information , which aims to adaptively fine-tune the locations of entities in the vector space for the entity alignment similarity measure. Our proposed AGN model can be naturally extended to entity alignment datasets across multiple temporal knowledge graphs. We evaluate our proposed model via temporal knowledge graphs on public datasets and our newly proposed noisy dataset. We also demonstrate the advantages of the AGN model through extensive experiments, which achieves state-of-the-art performance on the temporal knowledge graph dataset.

Emotion-cause pair extraction (ECPE) is an emerging task in emotion cause analysis, which extracts potential emotion-cause pairs from an emotional document. Most recent studies use end-to-end methods to tackle the ECPE task. However, these methods either suffer from a label sparsity problem or fail to model complicated relations between emotions and causes. Furthermore, they all do not consider explicit semantic information of clauses. To this end, we transform the ECPE task into a document-level machine reading comprehension (MRC) task and propose a Multi-turn MRC framework with Rethink mechanism (MM-R). Our framework can model complicated relations between emotions and causes while avoiding generating the pairing matrix (the leading cause of the label sparsity problem). Besides, the multi-turn structure can fuse explicit semantic information flow between emotions and causes. Extensive experiments on the benchmark emotion cause corpus demonstrate the effectiveness of our proposed framework, which outperforms existing state-of-the-art methods.