Intelligent diagnosis of faculty structure in higher education institutions based on knowledge graph and large language models

Abstract

Faculty structure is a key indicator of a university’s core competence, but traditional diagnosis relies on expert experience, manual statistics and one-dimensional analysis, and suffers from data barriers, dimensional separation and generalized suggestions. This study builds an intelligent diagnostic framework for faculty structure that integrates a Knowledge Graph (KG) and a Large Language Model (LLM). Taking the teacher as the core entity, we build a faculty knowledge graph with 8 entity types, 12 relationship types and about 27,000 nodes, then design a retrieval-augmented diagnostic model that injects KG sub-graphs into the LLM prompt as reasoning evidence to jointly identify six-dimensional risks—age, gender, title, academic qualification, academic origin and discipline—each governed by a literature- and expert-grounded quantitative threshold. On 2,803 teachers from a provincial undergraduate university (2020–2025), the system reaches 89.3% accuracy and a macro-F1 of 0.876, 15.8 percentage points above the expert-rule baseline, and an ablation study confirms that knowledge grounding contributes the largest share. The model identifies that 21.7% of colleges face the dual risk of “age fault + academic inbreeding”, and the LLM-generated suggestions score 4.2/5 in a 5-expert blind evaluation, significantly above a knowledge-free control. The study provides a feasible methodology and a reusable paradigm for data-driven, precise faculty governance.