A Hybrid Knowledge-Enhanced Legal AI System for Construction Contract Disputes

Abstract

With rapid urbanization and expanding infrastructure, construction contract disputes are increasing in volume and complexity, challenging traditional adjudication. This study proposes a domain-specific legal artificial intelligence (AI) system for construction contract disputes via hybrid knowledge integration based on the retrieval-augmented generation (RAG) paradigm, integrating five core legal texts and 500 adjudication cases within a dual-engine architecture. The knowledge base encodes legal concepts, relations, and rules to enable structured semantic inference. The DeepSeek-R1 reasoning engine analyzes case facts and legal logic via constrained generation, while the BGE-M3 retrieval module matches legal provisions and precedents using multivector indexing. A tripartite evaluation framework—semantic similarity, legal provision citation accuracy, and issue prediction F1 score—validates system performance. The hybrid knowledge model outperforms single-source models, achieving scores of 0.736, 0.952, and 0.937, respectively, while significantly reducing judicial document generation time. This study offers a theoretical and empirical basis for legal AI in Chinese construction disputes, demonstrating how integrating diverse legal knowledge enhances intelligent judicial assistance within China’s jurisdiction. It also provides a scalable methodological reference for the advancement of smart justice, with explicit recognition of its current jurisdictional limitations.