Deep Learning-Based Classification of Conference Paper Reviews: Accept or Reject?

Peer review plays a crucial role in ensuring that papers published in conferences are as true, valid, and accurate as possible. As part of the publication process, the scholarly work of an academic author undergoes review by domain experts, reflecting the commitment to uphold integrity. With the growing volume of scholarly submissions at the conference, the classification of the paper reviews through an automated classification system has become a pressing requirement. The primary aim of this research is to develop an automated classification system for conference paper reviews into two categories: accept and reject. However, the objective of this paper is twofold: developing an automated classification system and explaining the decision-making process through explainable artificial intelligence (XAI). First, to develop an automated classification system, we present a deep learning-based framework. Next, to explain the results produced by deep learning (DL) models, Local Interpretable Model-Agnostic Explanations (LIME), SHapley Additive exPlanations (SHAP), and Anchor Explainer have been applied. Among the applied DL techniques, the Bi-GRU-LSTM-CNN model attained the highest accuracy of 95.33% which indicates that the proposed methodology is able to replicate the human decision-making process in an effective manner.