Research on Automatic Construction Algorithm of Ideological and Political Knowledge Graph in Universities Based on Large Language Model

Abstract

Addressing the core pain points in current university ideological and political knowledge graph construction—poor domain adaptability of traditional methods, insufficient entity-relation extraction accuracy, weak implicit knowledge mining capabilities, and the difficulty of general-purpose large models adapting to the rigorous knowledge system and semantic norms of the ideological and political field—this paper proposes a hierarchical entity-relation joint extraction and knowledge calibration algorithm (LLM-HERCKA) driven by a large language model for the ideological and political field. First, a dedicated corpus covering four core ideological and political courses in universities and a fine-tuning paradigm are constructed to achieve deep adaptation of the large model to the ideological and political field. Second, a three-layer ontology architecture for ideological and political knowledge and a hierarchical joint extraction mechanism are designed to solve the error propagation problem of traditional pipeline methods. Finally, a dual-path knowledge calibration module is proposed to ensure the compliance and factual accuracy of extracted knowledge. Experiments were conducted on a self-constructed dataset of 2000 labeled samples for ideological and political education evaluation. Results show that the algorithm achieves an F1 score of 94.27% for entity extraction and 92.58% for relation extraction, representing improvements of 8.35 and 9.76 percentage points respectively compared to the traditional BERT+CRF model, and improvements of 5.12 and 6.34 percentage points respectively compared to the zero-sample extraction of general large-scale models. Ablation experiments validated the effectiveness of each core module. This algorithm can achieve efficient and automated construction of ideological and political knowledge graphs in universities, providing core technical support for digital teaching and intelligent applications of ideological and political education.