A knowledge graph-based large model training method and device, medium and equipment

Abstract

The application provides a knowledge graph-based large model training method and device, medium and equipment, the method comprises the following steps: acquiring a knowledge graph to be merged into a large model, and extracting a target knowledge path according to a graph traversal rule, the target knowledge path comprises one or more triples composed of a first entity, an inference relationship and a second entity; inputting the target knowledge path into a question and answer pair generation model to obtain a target question and answer pair; inputting the target question and answer pair into a question and answer pair scoring model to score in a language preference dimension by the question and answer pair scoring model to obtain a target score. The target training sample formed based on the target question and answer pair and the target score captures the structural information in the knowledge graph and clarifies the language preference tendency, so that the large model can effectively utilize the structured information of the knowledge graph and learn how to generate text more in line with the language preference in the training process based on the target training sample, thereby improving the reasoning ability and expression ability of the large model.