Automatic Literature Review Construction from Academic Papers Using Bidirectional and Auto-Regressive Transformers (BART)

Abstract

In the academic and research world, a literature review is an important stage in the research process. The abundance of scientific articles from various sources makes the literature review process time-consuming. Therefore, an automated system capable of efficiently summarizing scientific articles is needed. This research develops an automatic summarization system using the Bidirectional and Auto-Regressive Transformers [BART] method. This system is designed to identify the structure of scientific articles based on the IMRAD format [Introduction, Methods, Results, and Discussion] and produce more systematic and relevant summaries. The evaluation results using the BERTScore metric show that the developed model is capable of producing summaries with a precision level of 84.72%, a recall of 86.72%, and an F1-score of 85.69%. Thus, this system is expected to assist researchers in conducting literature reviews more quickly and effectively.