A Sentiment Augmented Deep Architecture to Predict Peer Review Outcomes

Peer review texts reflect the overall impression of the reviewers towards a candidate research paper and by far are the most important? artifact used by editors and program chairs to determine the prospective inclusion of a manuscript in a given journal or a conference. Here in this work, we study how we could make use of the sentiment information embedded within peer review texts to help editors or program chairs to make better editorial decisions. We design an efficient deep neural architecture that takes into account: the paper, the corresponding reviews, and sentiment polarity of the reviews to predict the recommendation score of reviewers and well as to anticipate the final decision. Our results show that we achieve significant improvement over the baselines (~ 29% error reduction) proposed in a recently released dataset of peer reviews.