BBPM: A Study of Information Pre-retrieval Models Based on Medical BERT Model

This study focuses on the preprocessing of medical paper retrieval, aiming to address the challenge of extracting information from lengthy medical texts. This process is crucial in facilitating effective medical research. We propose BioBERT Preprocessing Model (BBPM). By employing the cutting-edge medical BERT model, we predict the subject words of the paper’s abstract. Subsequently, based on these subject words, we calculate secondary similarity using the corresponding medical model. Finally, we incorporate the most strongly associated words to augment the associative vocabulary of the article, thereby enhancing its effectiveness in subsequent text retrieval. Experimental results show that BioBERT shows superior performance in predicting subject words in long medical texts, with a basic text similarity of about 90%. The primary contribution of this paper lies in the integration of the latest BioBERT model for preprocessing the abstracts of medical papers. This yields additional subject terms necessary for retrieval. These terms are then integrated with the title keywords into the final retrieval system, leading to more efficient retrieval outcomes. This approach promises to provide fresh insights into medical information retrieval.