Category detection method for field of evidence-based medicine

Abstract

The invention discloses a category detection method for the field of evidence-based medicine. The category detection method comprises the following steps that ELMo and Bi-LSTM processing are respectively carried out on each sentence in an abstract to obtain a sentence vector; encoding the sentence vector to obtain a text representation vector containing a semantic relationship between sentences; and inputting the text representation vector into a CRF model for sentence sequence classification, taking a to-be-classified sentence and a sentence category label as an observation sequence and a state sequence of the CRF model respectively, and obtaining a label probability of each sentence through sentence association characteristics extracted by a lower-layer network. According to the invention, evidence-based medical text abstract category detection is realized, and the multi-connection Bi-LSTM network is utilized to capture the dependency relationship and context information between sentences, the overall quality of sentence coding is improved in combination with a multi-layer self-attention mechanism, and a good effect is achieved on a public medical abstract data set.

Description

technical field [0001] The invention relates to the technical field of information processing of English medical text abstracts, in particular to a category detection method oriented to the field of evidence-based medicine. Background technique [0002] Evidence-Based Medicine (EBM) is an approach to clinical practice that obtains evidence by analyzing large medical literature databases such as PubMeb and searching texts on relevant clinical topics. EBM starts with a paper and further refines the evidence base on which a particular question rests through human judgment. The definition of clinical practice problems in the field of EBM often follows the PICO principle, namely: Population (P); Intervention (I); Comparison (C); Outcome (O). [0003] In order to complete the conversion from articles to medical evidence, in-depth combing of article abstracts is required. The abstract is a short statement without annotations and comments on the content of the medical article. It ...

AI Technical Summary

Problems solved by technology

[0010] When the existing work is used for the detection of clinical medical class labels, sentences are often classified separately, and the dependencies between words and sentences are not considered at the level of text expression, which will lead to poor classification results

Song et al. spliced ​​the overall encoding of the context of the sentence and the sentence vector to be classified for drug classification, lacking inter-sentence dependencies

Images