From reviews to decisions: A joint multitask aspect sentiment leveraged framework for assisting decision prediction from academic peer reviews

Peer reviews play a significant role in the quality of research articles published in prestigious venues, ensuring truth, validity, and originality. Predicting the outcome of a paper based on peer reviews is a daunting task, even for humans, irrespective of the number of dimensions and factors considered. This workload is already considered to be overburdening for the academic community. Based on the feedback from human reviewers, it is evident that Artificial Intelligence (AI) techniques may assist the editor/chair in anticipating the final decision. A peer review text reflects the reviewers’ opinions and sentiments about various aspects of the paper (e.g., novelty, substance, soundness, etc.) relevant to the proposed research. These aspects can serve as a basis for predicting the manuscript’s future (acceptance or rejection). In this study, we investigate how aspects and their corresponding sentiments can be leveraged to develop a multitask system that assists editors and chairpersons in determining manuscript outcomes, improving the editorial decision process. We conduct our experiments on the Aspect-enhanced Peer Review (ASAP) Review dataset. Experimental results show that our model achieves up to 81% accuracy in predicting the acceptance and rejection of a manuscript.