Research And Application Of Medical Entity Extraction Based On Multi-task Learning And Transfer Learning

With the promulgation and continuous promotion of national policies such as Healthy China2030 and the 14 th Five-Year Plan,intelligent medical treatment integrating artificial intelligence has become an important aspect in the digitalization and intelligent transformation of China’s medical and health industry.As medical and health care is closely related to people’s lives and depends on experience and knowledge,it requires accurate,comprehensive and high-quality knowledge organization and knowledge services.It is worth noting that medical entity extraction plays a vital role in the medical and health field,and is an important cornerstone of information extraction and knowledge organization.This specifically refers to the identification and extraction of various medical entities containing fine-grained medical knowledge and medical value in medical texts.In today’s era of explosive growth of medical and health big data,an efficient and accurate entity extraction method is urgently needed to realize the transformation and storage of unstructured medical knowledge,providing strong support and guarantee for intelligent question answering,auxiliary diagnosis and treatment,health management and other medical services.Despite its importance,medical entity extraction faces numerous difficulties and challenges due to the highly specialized and complex nature of the medical field.On the one hand,Chinese medical texts often contain a large number of modifiers and nested entities,making it challenging to define the boundaries of medical entities.Moreover,the lack of a space-like display separator in the Chinese language further complicates this issue.On the other hand,a significant number of medical entities are translated from foreign languages and contain English characters and vocabulary,resulting in various ways of expressing Chinese medical entities and further increasing the difficulty of extraction.Furthermore,medical entities are highly specialized and manually labeling them is a costly process,resulting in a scarcity of high-quality medical labeling data sets in China.Meanwhile,the number of unlabeled non-structured or semi-structured medical-related corpora is growing,making it challenging to effectively utilize the existing and growing medical data resources.As a result,addressing the problem of cold start for medical entity extraction data has become an important research focus in the field of medical natural language processing.To address the challenges and problems of medical entity extraction in the Chinese medical and health field,a novel approach that integrates multi-task learning and transfer learning within a deep learning model has been developed in this study.The method aims to leverage the benefits of both multi-task and transfer learning to improve the accuracy of medical entity extraction.Firstly,a hybrid deep learning entity extraction model was constructed using BERT,Bi LSTM,IDCNN and CRF to obtain word vectors,extract medical semantic features and output results.Transfer learning was applied through case,model,and feature transfer to enrich the semantic features and effectively use the medical data resources.At the same time,multi-task learning was employed to construct a coarsegrained tripartite task to improve entity boundary information sharing and medical entity definition.Furthermore,the model was optimized by introducing the self-attention mechanism and Highway network,resulting in the creation of the TLMT-BBIC-HS medical entity extraction model,which was used to develop a medical health question-answering system.The study’s results indicate that the TLMT-BBIC-HS achieved an F1 value of 92.98% in the Chinese medical entity extraction dataset,which is 15.99% and 16.44% higher than the benchmark models BERT-Bi LSTM-CRF and BERTIDCNN-CRF,respectively.This significant improvement of the model’s accuracy provides strong support for health knowledge mapping and question-and-answer systems in the medical and health field.This study is helpful to further promote the development of knowledge organization and knowledge service in the medical and health field,and has important significance for the application and deepening of artificial intelligence in intelligent medicine.