Analyzing we-sentences in Scientific Writing to Predict Peer Review Outcomes

Abstract

Peer review is one of the mainstay tasks of scientific publishing. Nonetheless, the high volume of submissions, potential bias in peer review, and time dedicated to this process are problems related to publishing. Such challenges have inspired the development of computational methods aimed at automating publishing decisions in the peer review process. In this context, we studied the value of the information within phrases containing first-person plural pronouns (we-sentences) in English benchmark datasets of accepted and rejected scientific papers as a factor to differentiate them. Afterward, the information in these phrases was leveraged from a text classification perspective by applying two approaches: 1) training with we-sentences only and 2) using term weighting schemes to generate a text representation that emphasizes the value of the words in these sentences. Our results confirm that we-sentences express essential information about the content of the paper (e.g., findings, methods, contributions, and conclusions). Therefore, these sentences embed relevant lexical patterns that contribute to determining the final acceptance/rejection decision. Finally, our findings can be helpful in developing tools to assist the review process of research papers or as a source of feedback for authors.