An LLM-enhanced intelligent evaluation framework with knowledge graph integration and path reasoning
Abstract
This paper addresses the challenges of identifying instability and the lack of explicit reasoning grounds in the processing of complex engineering texts by proposing an intelligent algorithm system tailored for review tasks. Existing methods primarily rely on pure text semantic modeling and struggle to effectively utilize structured constraint information, resulting in insufficient consistency and reliability of results. To address this, we design a computational pipeline comprising “structured extraction—knowledge enhancement—path reasoning.” The system first utilizes large language models to perform triplet extraction and achieves entity normalization through similarity constraints to construct a consistent knowledge graph. Subsequently, it obtains relevant subgraphs through entity matching and neighborhood expansion, incorporating structured knowledge into joint modeling. Furthermore, by combining a path scoring function to filter candidate inference chains, the system achieves query-oriented optimal path selection and drives the result generation process. Experimental results demonstrate that this method achieves significant improvements in review decision-making tasks, with an accuracy of 0.8667 and a triplet extraction F1 score of 0.9600, reflecting strong semantic modeling and structural integration capabilities. In summary, under a unified computational workflow, this system achieves effective synergy between structured knowledge and semantic modeling, balancing classification performance with reasoning traceability, and demonstrates good generalizability and potential for engineering applications.