Scholarlink

<p>Managing and retrieving complex legal information poses significant challenges, such as handling vast, interconnected legal texts efficiently and ensuring precise interpretation based on context, jurisdiction, and precedents. This paper addresses a critical gap in legal information retrieval systems by introducing an integrated knowledge graph-based approach combined with agent-driven decision making, with Tamil language support as an accessibility feature for regional users. Our comprehensive analysis of existing literature reveals that while current approaches address isolated aspects of the problem, none provides a holistic dynamic solution. Our novel framework introduces two primary contributions: (1) a hierarchical knowledge graph schema that enables structured legal reasoning through multi-layered entity relationships, and (2) adaptive agent reasoning capabilities that dynamically navigate complex legal knowledge structures for contextually relevant responses. The system architecture integrates these core innovations with context-aware graph-based retrieval-augmented generation (RAG) for precise retrieval and incorporates Tamil language processing to enhance accessibility for regional users. Additionally, our approach demonstrates the significance of intent identification fine-tuning, which improves query understanding precision by 23% compared to baseline approaches. Experimental results demonstrate significant improvements in retrieval accuracy and query comprehension compared to traditional keyword-based approaches. Our hierarchical knowledge graph achieved a modularity score of 0.646 using Louvain community detection, while our LangGraph agent attained ROUGE-1/2/L scores of 0.467/0.311/0.508 for single-document tasks. For more complex multi-document retrieval, the ReAct Agent achieved impressive scores of 0.592/0.432/0.512, with high semantic similarity scores of 0.831 and 0.893. The proposed system offers a scalable solution that leverages structured knowledge representation and adaptive reasoning to navigate complex legal relationships, with Tamil language support providing enhanced accessibility for regional users in the Indian legal domain.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hierarchical knowledge graph based legal information retrieval from multiple documents using intelligent agents

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • Durairaj Thenmozhi,
  • Makesh Vaibhav AG,
  • Naren Srinivasan M

摘要

Managing and retrieving complex legal information poses significant challenges, such as handling vast, interconnected legal texts efficiently and ensuring precise interpretation based on context, jurisdiction, and precedents. This paper addresses a critical gap in legal information retrieval systems by introducing an integrated knowledge graph-based approach combined with agent-driven decision making, with Tamil language support as an accessibility feature for regional users. Our comprehensive analysis of existing literature reveals that while current approaches address isolated aspects of the problem, none provides a holistic dynamic solution. Our novel framework introduces two primary contributions: (1) a hierarchical knowledge graph schema that enables structured legal reasoning through multi-layered entity relationships, and (2) adaptive agent reasoning capabilities that dynamically navigate complex legal knowledge structures for contextually relevant responses. The system architecture integrates these core innovations with context-aware graph-based retrieval-augmented generation (RAG) for precise retrieval and incorporates Tamil language processing to enhance accessibility for regional users. Additionally, our approach demonstrates the significance of intent identification fine-tuning, which improves query understanding precision by 23% compared to baseline approaches. Experimental results demonstrate significant improvements in retrieval accuracy and query comprehension compared to traditional keyword-based approaches. Our hierarchical knowledge graph achieved a modularity score of 0.646 using Louvain community detection, while our LangGraph agent attained ROUGE-1/2/L scores of 0.467/0.311/0.508 for single-document tasks. For more complex multi-document retrieval, the ReAct Agent achieved impressive scores of 0.592/0.432/0.512, with high semantic similarity scores of 0.831 and 0.893. The proposed system offers a scalable solution that leverages structured knowledge representation and adaptive reasoning to navigate complex legal relationships, with Tamil language support providing enhanced accessibility for regional users in the Indian legal domain.