Text Classification Algorithm for Medical Adverse Events Based on Deep Learning

The management of medical adverse events is an important topic in the field of healthcare. The classification of medical adverse events is the primary link in adverse event management. However, the research on medical adverse event classification technology is still in its early stages, mainly using rules or machine learning methods, and the prediction accuracy is not high enough. To address the above issues, this paper explores the application of deep learning in medical adverse event text classification tasks and proposes a hybrid model that combines deep pyramid convolution and cyclic convolution network structures to achieve category classification of adverse events. This model has two feature extraction modules: deep pyramid convolution and cyclic convolution. These two modules extract features in parallel using BERT word vectors as inputs, and then concatenate the features. The experiment shows that the model has good performance in adverse event classification. In order to further improve the effectiveness of adverse event text classification, this paper proposes two improved models, BRDCN and BRDHW, based on the hybrid model. The BRDCN model abstracts the concatenated features by adding a convolutional network after the feature extraction module to further extract features from the concatenated tensor. The experiment shows that the multi classification performance of adverse event cate-gories in the BRDCN model has improved compared to before improvement. The BRDHW model refers to the gate control structure of the highway network and sets a gate control matrix. This matrix is used to fuse the features extracted by the two modules. The experiment shows that the BRDHW model's multi classification performance of adverse event categories has been further improved.