Named entity recognition method for Chinese medical records

Abstract

The invention discloses a named entity recognition method for Chinese medical records, and relates to the technical field of natural language processing. The method comprises the steps of carrying outthe preprocessing of a data text; pre-training the character vector model and the word vector model; converting the text sequence into word-based vector representation, introducing a word vector to obtain a mixed vector, and encoding the mixed vector to obtain an encoded text feature vector sequence; inputting the encoded text feature vector sequence, and obtaining an attention vector sequence byusing a multi-head attention mechanism; respectively inputting the attention vector sequences into two layers of CRF classifiers to obtain final label output; updating and converging the model; and performing named entity identification on the Chinese medical records. According to the method, a WC-LSTM network structure of a self-attention mechanism is introduced, and the finally output label isdetermined by using two-layer CRF classification based on the labels with different granularities, so that the time consumed by training is shortened while the named entity recognition accuracy in thefield of Chinese medical records is improved.

Description

technical field [0001] The invention relates to the technical field of natural language processing, in particular to a named entity recognition method for Chinese medical records. Background technique [0002] With the development of artificial intelligence technology, especially natural language processing technology, Chinese electronic medical records, which have extensive accumulation and imply huge potential information, have begun to attract people's attention. Named entity recognition refers to the positioning and classification of named entities mentioned in unstructured texts into predefined categories. The field of Chinese medical records used in the present invention is generally defined as: disease and diagnosis, inspection, inspection, surgery, medicine and anatomy parts. Named entity recognition is the cornerstone of many natural language processing technologies, such as knowledge graphs, relationship extraction, etc., and is the first step in using artificial ...

AI Technical Summary

Problems solved by technology

[0005] 1. The word segmentation information of each sentence is different, so the network structure is not fixed for each training, and batch training cannot be performed, so the training takes a lot of time

[0006] 2. In Chinese medical records, various labels often overlap. For example, "cardia and gastric fundus cancer" is a disease-type entity, but "cardia" and "stomach" are anatomical entities. This method does not take this situation into account , making part of the label output inaccurate

Images