Named Entity Recognition of Electronic Medical Records Based on Multi-Feature Fusion

Abstract

Named entity recognition (NER) is a very basic task in natural language processing (NLP). The paper studies the problem of named entity recognition in Chinese electronic medical records, and proposes a method based on the Bert-Bi-LSTM-CRF model. In addition, the model incorporates the functionality of radical components and dictionaries to improve the recognition accuracy. The complexity of Chinese medical record entities, the ambiguity in language expression, and the lack of adequate labeled data make traditional rule-based or machine learning methods less effective. To address this problem, we adopt the Bert-Bi-LSTM-CRF model, which effectively captures contextual information and semantic relationships to improve entity recognition accuracy. Furthermore, to further enhance the model's performance, we introduce the functionality of radical components and dictionaries. Radical components are an important component of Chinese characters and can be used to assist in identifying entities and improve the model's generalization ability. We also utilize medical dictionaries to assist in entity recognition. These dictionaries contain rich medical terms and vocabulary, which can effectively help the model identify entities. The proposed method is evaluated on the public dataset CCKS2019, and the experimental results demonstrate that it outperforms traditional methods, achieving an F1 score improvement of nearly 7 percentage points, and achieves good experimental results.