Beyond Binary Edges: How Hyperedge-Structured Knowledge Graphs Eliminate Clause Fragmentation in LLM-Driven Contract Attribute Extraction

Large Language Models have demonstrated remarkable capability in contract analysis, yet they consistently fail at a fundamental task: maintaining clause integrity across multi-party, cross-referenced legal provisions. This paper identifies the root cause as the binary-edge limitation of conventional knowledge graph representations, which forces complex N-ary legal relationships into lossy pairwise decompositions. We introduce a hyperedge-structured knowledge graph architecture that preserves the natural N-ary structure of legal clauses, enabling LLMs to extract contract attributes without the fragmentation artifacts that plague current approaches.

Experimental results across commercial contract corpora demonstrate significant improvements in extraction accuracy, particularly for indemnification clauses, multi-tiered obligation structures, and cross-referenced condition precedents.