CMLNER: Chinese medical literature named entity recognition based on
multi-feature fusion and attention mechanism
Abstract
Background: With the rapid development of big data and
artificial intelligence, the medical knowledge in medical literature has
attracted more and more attention from the academic community. As the
foundation for medical knowledge extraction and decision support system
construction, Chinese medical literature named entity recognition
(CMLNER) is the process of automatically recognizing entities in medical
literature. Due to the diverse types, unclear boundary, complex
composition and lack of explicit separators like space in Chinese
medical entity, the task of CMLNER is more complicated compared with
English medical literature named entity recognition. Objective:
In this study, we aim to investigate novel methods to model CMLNER and
analyze the results. Methods: This study proposes a novel
neural network architectural model MFA-BERT-BiLSTM-CRF based on external
medical knowledge and self-attention mechanism. Results:
Compared with traditional NER methods, the proposed model could more
effectively capture rich medical semantic features, global context
information and further improve the performance of CMLNER. In addition,
the key factors affecting CMLNER is also analyzed using the correlation
coefficient method and the result indicates that the number and
composition rules of entity are main factors. Finally, the recognition
performance and error results are also analyzed in this paper.
Conclusions: Our research makes up for the deficiency of
previous frameworks and will further promote the development of medical
entity recognition.