Chinese Medical Named Entity Recognition Based on Pre-trained Language Models

Abstract

With the progressive development of academic research and the increasing standard of living, the need for medical entity identification is gradually increasing. Medical entity recognition has been rapidly developed to help build medical knowledge graphs and disease prediction, improve diagnostic accuracy, enable disease prevention, simplify clinical decision-making and reduce healthcare costs. In Chinese named entity recognition, the input is a word-level vector representation, but in Chinese text, words are the smallest units that express meaning. Although the BERT model has the advantage of excellent training results and can avoid the noise generated by the different word-level vector when the input word-level vector is used, it is wasteful for Chinese words that possess more information. Therefore, we propose a pre-training model based on Tag Embedding and Simple Lexicon word strengthening with word information fusion, which incorporates the word boundary information of the text into the more expressive text after encoding, increasing the amount of information it contains. The purpose of this study is to build a named entity recognition model using deep neural networks to improve the accuracy of medical-related entity recognition from the perspective of improving the information exploiting of medical data sets in the current era of big data informatics, and to contribute to related research.