More about HKUST
A Survey on LLM-Based Multi-Agent Systems for Document Mining
PhD Qualifying Examination Title: "A Survey on LLM-Based Multi-Agent Systems for Document Mining" by Mr. Pengze CHEN Abstract: Document mining is the process of extracting valuable information and knowledge from large collections of documents, such as academic literature, analysis reports, and medical records. Efficient document mining is crucial for expediting the collection, processing, and utilization of massive documents in this era of unprecedented data growth. However, traditional methods have often been limited by a shallow understanding of semantics, being largely confined to keyword matching and statistical patterns. The true potential of document mining is only now being unlocked with the advent of Large Language Models (LLMs). Their capabilities in semantic understanding, reasoning, and multimodal processing empower them to comprehend and interact with documents in a human-like manner. Furthermore, the emergence of LLM-based multi-agent systems (LMASs) enhances this process through collaborative execution and mitigation of hallucinations, significantly boosting the reliability and effectiveness. To systematically investigate the advancements in this field, this survey organizes the landscape around three representative capabilities: document retrieval, for precisely locating relevant documents; document answering, for delivering direct answers; and document summarization, for creating condensed and coherent summaries of documents. It first analyzes the requirements and challenges of each task, and subsequently examines the existing solutions, highlighting their respective strengths and weaknesses. Finally, it summarizes the existing works and outlines promising research directions to further advance LMAS for document mining. Date: Monday, 29 September 2025 Time: 3:00pm - 5:00pm Venue: Room 5501 Lifts 25/26 Committee Members: Prof. Lei Chen (Supervisor) Prof. Ke Yi (Chairperson) Dr. Xiaomin Ouyang