Expertise Computation for Automatic Reviewer Assignment

Divya Kwatra, Vasudha Bhatnagar

Abstract: Assigning effective and accurate reviewers is a crucial step in the peer-reviewing process to ensure high-quality publications. Automatic reviewer assignment problem, which aims to assign experts for rating the submitted research, relies critically on determining the expertise of the reviewers in the topics covered in the manuscript and assigning them so that all research topics covered in the manuscript are rated by proficient reviewers. In this paper, we consider research expertise as a commixture of interest and proficiency of the reviewers in the topics that span the manuscript. We take reviewers’ recent publications and self-declaration about the areas of research interest as input. Lexical variations in the topics declared by reviewers are leveled using sentence transformers and the research publications. Similarity between the semantic content of the manuscript and the research topics is also matched using the sentence transformer framework. Subsequently, we quantify the reviewers’ interest in each topic along with their proficiency in that topic based on the number of authored publications. Reviewers are scored for expertise, which is a function of the two computed quantities. Top scoring reviewers are assigned the manuscript for review ratings. We evaluate the effectiveness of the proposed method using topic coverage and review confidence as metrics. We observe that the proposed algorithm for expertise computation and reviewer assignment performs better than the baseline method.