Biomedical text-level relation extraction and judgment method based on agent workflow

Abstract

The application discloses a biomedical text-level relation extraction and judgment method based on an agent workflow, comprising the following steps: constructing a relation extraction workflow, preprocessing biomedical text by using a large language model to generate structured sentences, extracting relation triples from a syntax tree through dependency grammar analysis, and storing the relation triples in a knowledge base after merging and removing duplicates and injecting a document identifier; constructing a relation judgment workflow, retrieving relevant triples as evidence from the knowledge base when a relation query statement containing a search target is input, inputting the large language model for reasoning, and finally outputting an interpretable judgment result based on the evidence; the method realizes a complete workflow from automatic relation extraction of text to supportable interpretable query judgment, and can efficiently process large-scale biomedical literature data.

Images