Self-supervised extractive text summarization for biomedical literatures

In this study, we propose a self-supervised approach to extractive text summarization for biomedical literature. The approach uses abstracts to find the most informative content in the article, then generate a summary for training a classification model. The Sentences in the abstract and literature were first embedded using BERT. A similarity-based model was then applied to label the informative sentences for training the classifier. We used logistic regression as our classification model and used the features of sentence embedding for the classification. The results showed the feasibility of employing the abstract to perform self-supervised training of a classification model to generate extractive summarization. This approach can enable automatic generation of one or two-page executive summaries of biomedical literature to keep clinicians and biomedical researchers up to date with the latest development