Multi-task question and answer driven medical entity relationship extraction method

Abstract

The invention discloses a multi-task question and answer driven medical entity relationship extraction method, which comprises the following steps of: by giving a symptom description text and an entity type set, constructing entity extraction question query description for each entity type; constructing a multi-layer semantic representation model; constructing a prediction model to obtain a starting index set and an ending index set; pairing the index numbers of the starting index set and the ending index set in pairs, performing matching prediction on each starting position and each ending position by constructing a prediction model, and extracting entities matched with the starting positions and the ending positions to obtain a plurality of candidate entities; determining the types of the candidate entities by constructing a multivariate classifier to obtain a candidate entity set; constructing a relation extraction problem description between any two entities in the candidate entity set, and splicing the relation extraction problem description with the original medical record description to form a new sentence; and predicting the relation type of every two entities one by one by learning the context semantic representation of the whole new sentence and connecting a full connection layer. According to the method, a reading understanding mechanism is introduced when the problems of word ambiguity and nesting of entities in electronic medical records are solved.

Description

technical field [0001] The invention relates to the field of text information extraction, in particular to a method for extracting medical entity relations driven by multi-task questions and answers. Background technique [0002] With the development of medical informatization, a large amount of medical data has been generated. Electronic medical records record the detailed diagnosis and treatment process of patients. A large amount of medical record data plays an important role for follow-up medical researchers. How to quickly and accurately extract key data in medical record texts has become important means. [0003] Most of the existing extractions of medical entities and their relationships are based on grammar. For example, Chinese patent document CN112883736 discloses a method and device for extracting medical entity relationships. Based on the BERT part of the joint extraction model, entities are extracted from medical electronic medical records. Obtain each entity c...

AI Technical Summary

Problems solved by technology

[0004] Existing methods rely on syntactic dependency analysis and use convolutional neural networks and recurrent neural networks to capture sentence representations. These methods are difficult to solve the problem of word ambiguity and nesting in entities in electronic medical records, resulting in inaccurate extraction of medical entities

Images