WKWenjun Ke

Wenjun Ke

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Wenjun Ke has a total of 28 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • LogicICL: Towards Instructive Logical-Aware Demonstrations for Stimulating LLMs Mathematical Reasoning Ability

    by Yao He, Jiajun Liu, Wantao Wen, Yikai Guo, Hanchao Chen, Yue Ma, Wenjun Ke, Peng Wang

    Large language models (LLMs) have demonstrated remarkable capabilities on various downstream tasks. However, mathematical problem-solving remains a formidable challenge due to its complex reasoningsteps. To address these issues, in-context learning (ICL) has emerged as a training-free paradigm that enhances LLMs reasoning ability by retrieving demonstrations. Existing ICL methods can be broadly categorized into similarity-based retrieval (SSR) and structural similarity-based retrieval (STR) methods. SSR methods retrieve demonstrations based on the similarity of texts in the semantic space, yet they fail to identify the underlying mathematical reasoning structure. To overcome this limitation, STR methods retrieve examples by aligning resoning structures between the target and candidate problems. However, they still face two major challenges: (1) misalignment in ensuring consistent reasoning paths between retrieved demonstrations and the target problem, and (2) inability to identify the underlying theorems required to solve the target problem. To alleviate these challenges, we propose LogicICL, a logic-aware ICL framework that retrieves examples with consistent inference paths and formulaic alignment by explicitly modeling symbolic reasoning. Specifically, (1) to address the misalignment in reasoning steps, we translate all problems into first-order logic (FOL) representations and mine circuit-level units for structured comparison of inference processes. (2) to capture the underlying theorems, we build logic circuit graphs from these units and apply graph-based similarity to retrieve demonstrations with consistent symbolic derivation and formulaic alignment. Experimental results show that LogicICL outperforms existing state-of-the-art baselines on the seven benchmarks, achieving average accuracy gains of 2.48%, 1.30%, 1.66% and 3.88% on the four backbones, respectively.


  • Open Access

    Incorporating LLM Versus LLM Into Multimodal Chain-of-Thought for Fine-Grained Evidence Generation

    by Shuyuan Xing, Yao He, Hanchao Chen, Wenjun Ke

    Multimodal Chain-of-Thought (MCoT) has become an effective strategy for enhancing multimodal large-language models (MLLMs) by breaking down complex tasks into sequential reasoning steps. Despite its interpretability benefits, MCoT often encounters difficulties with fine-grained semantic grounding, particularly when reasoning involves small objects, subtle attributes, or visually complex scenes that can lead to inaccuracies. Existing attempts to address these issues primarily fall into two categories: fine-tuning, which depends on large annotated datasets and costly parameter updates; and in-context learning (ICL), which achieves few-shot or zero-shot reasoning without model modification. Although ICL provides flexibility and adaptability, it is prone to semantic drift caused by an unstable prompt quality. To overcome these limitations, this study presents an entity-level evidence generation and verification framework using the ICL paradigm. This approach first produces MCoT from multimodal inputs, followed by extraction of key entities with enriched evidential descriptions. These entities were then cross-validated through adversarial checks using multiple MLLMs, and the verified evidence was integrated back into the reasoning chain. Experiments demonstrated consistent performance gains: on ScienceQA, the accuracy improved from 82.39% to 86.04%(+3.65%) with GPT-3.5, 84.96% to 89.37%(+4.41%) with Gemini; on MathVista, the accuracy increased from 43.1% to 43.6%(+0.50%) with GPT-3.5, and from 44.7% to 45.6%(+0.90%) with Gemini. These results establish newstate-of-the-art baselines and confirm the robustness and generalizability of the entity-level verification for multimodal reasoning.


  • Open Access

    Large Language Models in Document Intelligence: A Comprehensive Survey, Recent Advances, Challenges, and Future Trends

    by Wenjun Ke, Yifan Zheng, Yining Li, Hengyuan Xu, Dong Nie, Peng Wang, Yao He

    The rapid proliferation of documents has made document intelligence increasingly critical across various industries. In recent years, Large Language Models (LLMs) have dramatically transformed the field of document intelligence, allowing for more advanced and accurate document processing solutions. Despite these advancements, most existing surveys have failed to focus on these breakthroughs, instead concentrating on traditional methods and earlier machine learning techniques. This survey seeks to fill that gap by offering an in-depth analysis of approximately 300 papers published between 2021 and mid-2025, thus providing a comprehensive overview of the impact of LLMs in document intelligence. The key topics explored include Retrieval-Augmented Generation (RAG), long-context processing, and fine-tuning LLMs for document comprehension. Furthermore, the survey highlights essential datasets, practical applications, current challenges, and future research directions, offering critical insights for both researchers and industry practitioners looking to advance the field.


  • Learning Multi-Granularity and Adaptive Representation for Knowledge Graph Reasoning

    by Ziyu Shang, Peng Wang, Jianghan Liu, Jiajun Liu, Guozheng Li, Zijie Xu, Zhizhao Luo, Xiye Chen, Wenjun Ke

    Knowledge graph reasoning (KGR) seeks to infer new factual triples from existing knowledge graphs (KGs). Recent methods have unified transductive and inductive reasoning by learning entity-independent representations through local neighboring structures. Nevertheless, these methods often encounter inefficiencies and rely on elaborate local structures without directly modeling the correlations between queries and various structures within KGs. In this paper, we propose a novel framework MulGA, which is designed to learn multi-granularity and adaptive embeddings for KGR. MulGA first employs connectivity subgraphs to uniformly and hierarchically represent query-related structures within KGs, such as triples, relation paths, and subgraphs, establishing the hierarchical relationship between structures at different granularities. Subsequently, we design a graph neural network-based multi-granularity embedding propagation module that unifies the message-passing process with the connectivity subgraph construction. This module obtains the query-related structural representations by all entities at multiple granularities, eliminating the need to explicitly extract any graph elements, thus addressing inefficiency issues. Moreover, we develop a structure-aware adaptive merging mechanism that assigns weights to different granularities and integrates them into cohesive subgraph-granularity representations for reasoning. The systematic experiments have been conducted on 15 benchmarks and MulGA achieves a significant improvement in MRR by an average of 0.5% -1.1% on transductive tasks and 0.2% -7.3% on inductive tasks than existing state-of-the-art methods. Moreover, MulGA exhibits faster convergence speed, smaller number of parameters, competitive inference time, and alleviates the over-smoothing prevalent in graph neural networks.


  • Open Access

    Techniques for Fine-Grained Analysis of Scientific and Technological Intelligence

    by Xiaoyan Zhao, Yao He, Yankun Gao, Xu Luo, Wenjun Ke

    In today's world, scientific and technological intelligence information presents challenges such as insufficient annotated samples in interdisciplinary and specialized fields, the continuous emergence of new technologies, and complex interwoven dependencies of technical points. These aspects pose significant obstacles for the fine-grained acquisition and analysis of scientific and technological intelligence. To address these issues, this paper focuses on three areas: domain-aware generation of scientific and technological intelligence samples, extraction of intelligence entity relationships based on continual learning, and mining of technological dependency chains using causal graphs. This research aims to provide foundational techniques and algorithms for the acquisition and analysis of fine-grained scientific and technological intelligence, thereby enhancing our nation's capability in technological trend assessment and development planning in the context of global competition.