Designing a Hybrid Knowledge Graph–Large Language Model Framework for Construction Contract Analysis: A Design Science Research Approach

Abstract

Construction contract analysis increasingly relies on combining Knowledge Graphs (KGs), which provide structured, interpretable representations of contractual entities and relationships, with Large Language Models (LLMs), which offer flexible extraction and reasoning over unstructured clause text. Existing hybrid KG–LLM systems for contract analysis remain architecturally fragmented, are evaluated using narrow, inconsistent metrics, and depend heavily on manually maintained ontologies, limiting their scalability and reliability for construction practice. Following the Design Science Research (DSR) methodology of Peffers et al., this study analyzes 18 studies on KG–LLM integration in legal and construction-contractual settings to identify four recurring gaps: limited scalability of expert-curated knowledge structures, a context-fidelity gap in retrieval-augmented reasoning, inconsistent risk-mitigation strategies, and the absence of standardized, multi-dimensional evaluation criteria. Building directly on these findings, the resulting design artifact is a four-layer conceptual architecture comprising contract preprocessing, hybrid knowledge-graph construction, retrieval-augmented KG–LLM integration, and explainable reasoning, together with a proposed multi-dimensional evaluation protocol addressing knowledge quality, retrieval performance, reasoning traceability, legal consistency, and explainability. The architecture has not been implemented; its operational logic is illustrated through a worked, non-empirical construction-contract scenario, with empirical validation against real contract corpora identified as the necessary next step. This study contributes an evidence-grounded conceptual design intended to guide future implementations that advance the scalability, transparency, and reliability of hybrid KG–LLM systems for construction contract analysis.