Hybrid Deep Learning Based Model on Sentiment Analysis of Peer Reviews on Scientific Papers

The peer review process involved in evaluating academic papers submitted to journals and conferences is very perplexing as at times the scores given by the reviewer may be poor in contrast with the textual comments which are in a positive light. In such a case, it becomes difficult for the judging chair to come to a concrete decision regarding the accept or reject decision of the papers. In our paper, we aim to extract the sentiment from the reviewers’ opinions and use it along with the numerical scores to correlate that in order to predict the orientation of the review, i.e., the degree of acceptance. Our proposed methods include Machine learning models like Naive Bayes, Deep learning models involving LSTM and a Hybrid model with BiLSTM, LSTM, CNN, and finally Graph based model GCN. The dataset is taken from the UCI repository consisting of peer reviews in Spanish along with other parameters used for judging a paper. Bernoulli’s Naive Bayes was the model that fared the highest amongst all the approaches, with an accuracy of 75.61% after varying the parameters to enhance the accuracy.