Exploring Artificial Intelligence Techniques for Peer Review and Establishing a Research Lineage

Abstract

In the realm of academia, peer review has a longstanding tradition of enhancing the rigor of newlinescientific research. It involves researchers receiving valuable feedback from their peers, allowing newlinethem to refine their work and ensure that only the highest-quality research contributes to our newlinecollective knowledge. However, the peer review system faces significant challenges today. The newlinegrowing number of paper submissions and publication outlets has led to a scarcity of reviewers, newlinemaking it challenging to identify domain experts and maintain review standards. To tackle these newlineissues to some extent, we propose the integration of artificial intelligence (AI) techniques into newlinethe peer review process. This integration aims to enhance the efficiency and effectiveness of peer newlinereview while upholding its quality. This thesis focuses on enhancing scholarly communication by newlineutilizing Artificial Intelligence (AI) techniques to improve the peer review process. The central newlinegoals are to elevate the quality of review comments and enhance the transparency and efficiency newlineof peer review. In this thesis, we utilize machine learning (ML) and natural language processing newline(NLP) techniques to explore and address significant issues related to scholarly peer review. newlineIn our first contribution, we focus on leveraging the interactions within paper review texts to newlinepredict review decisions. By employing deep neural networks guided by cross-attention mechanisms, newlineour proposed models demonstrate promising performance in predicting review decisions. newlineImportantly, our aim here is not to replace human reviewers but to establish a human-AI collaboration newlinewhere AI systems analyze review texts and paper content to predict the fate of a paper. newlineTransitioning to our second contribution, we focus on automatically generating recommendations newlineand predicting decisions. This is accomplished by analyzing the full text of paper and corresponding newlinereview texts. We employ deep neural network models equipped with attention mechanisms, newlinesurpassing existing benchmarks. This provide