A deep learning approach for identifying biomedical breakthrough discoveries using context analysis

Abstract

AbstractBreakthrough research in scientific fields usually comes as a manifestation of major development and advancement. These advances build to an epiphany where new ways of thinking about a problem become possible. Identifying breakthrough research can be useful for cultivating and funding further innovation. This article presents a new method for identifying scientific breakthroughs from research papers based on cue words commonly associated with major advancements. We looked for specific terms signifying scientific breakthroughs in citing sentences to identify breakthrough articles. By setting a threshold for the number of citing sentences (“citances”) with breakthrough cue words that peer scholars often use when evaluating research, we identified articles containing breakthrough research. We call this approach the “others-evaluation” process. We then shortlisted candidates from the selected articles based on the authors’ evaluations of their own research, found in the abstracts. This we call the “self-evaluation” process. Combining the two approaches into a dual “others-self” evaluation process, we arrived at a sample of 237 potential breakthrough articles, most of which are recommended by the Faculty Opinions. Based on the breakthrough articles identified, using SVM, TextCNN, and BERT to train the models to identify abstracts with breakthrough evaluations. This automatic identification model can greatly simplify the process of others-self-evaluation process and promote identifying breakthrough research.