Entity Relationship Generation in the power Field based on Document-level

This study aims to enhance entity relationship extraction in the power field to improve knowledge management, predictive maintenance, and fault diagnosis. The paper proposes a document-level relationship extraction method using large language models (LLMs) to identify relationships between new and known entities in power standard knowledge documents. It generates documents through prompt chains, extracts relationship evidence, and constructs an entity relationship graph. The LLM is fine-tuned with historical power data to build a comparative knowledge graph, and a consistency denoising method is used to merge and clarify the graphs. Experiments on a Chinese power knowledge dataset show that the method effectively extracts accurate entity relationships, especially between new and known entities. This approach improves the precision of relationship extraction, offering both theoretical and practical benefits for managing and utilizing power knowledge.