Large model-based knowledge graph construction enhancement method

Abstract

The application discloses a knowledge graph construction enhancement method based on a large model, which comprises the following steps: a double-mechanism framework combining prompt learning and thought chain iterative verification is adopted to extract relation triples from field standard corpus, and through multiple rounds of iterative verification and correction, a reliable triple set is output; a hierarchical fusion architecture based on coarse-grained clustering and fine-grained semantic alignment is adopted to first perform coarse-grained semantic grouping on the reliable triple set, then utilize a large model to perform fine-grained semantic alignment and normalization processing on the triples in the group, and output a standardized triple set; a closed-loop mechanism comprising global graph input, large model reasoning, confidence screening and iterative optimization is constructed, an initial knowledge graph based on the standardized triple set is constructed, potential relations among entities are mined, the graph is dynamically expanded after confidence screening, and the knowledge graph is continuously and autonomously evolved through iterative optimization; the application does not require excessive manual intervention, is convenient to operate, and is highly scalable.

Images