A Category Detection Method for Evidence-Based Medicine

Evidence-Based Medicine (EBM) gathers evidence by analyzing large databases of medical literatures and retrieving relevant clinical thematic texts. However, the abstracts of medical articles generally show the themes of clinical practice, populations, research methods and experimental results of the thesis in an unstructurized manner, rendering inefficient retrieval of medical evidence. Abstract sentences contain contextual information, and there are complex semantic and grammatical correlations between them, making its classification different from that of independent sentences. This paper proposes a category detection algorithm based on Hierarchical Multi-connected Network (HMcN), regarding the category detection of EBM as a matter of classification of sequential sentences. The algorithm contains multiple structures: (1) The underlying layer produces a sentence vector by combining the pre-trained language model with Bi-directional Long Short Term Memory Network (Bi-LSTM), and applies a multi-layered self-attention structure to the sentence vector so as to work out the internal dependencies of the sentences. (2) The upper layer uses the multi-connected Bi-LSTMs model to directly read the original input sequence to add the contextual information for the sentence vector in the abstract. (3) The top layer optimizes the tag sequence by means of the conditional random field (CRF) model. The extensive experiments on public datasets have demonstrated that the performance of the HMcN model in medical category detection is superior to that of the state-of-the-art text classification method, and the F1 value has increased by 0.4%–0.9%.