Book Review Sentiment Classification in Bangla using Deep Learning and Transformer Model

Sentiment analysis is the approach detecting different underlying emotions from text data. E-commerce business is getting popular all over the world, including in Bangla-speaking regions. E-commerce websites provide different kinds of products and services. Following the growth of e-commerce businesses, there is a gap in research on consumer sentiment analysis. Our study focused on sentiment analysis of book reviews in Bangla. In this paper, we have developed our dataset of 5189 reviews by crawling data through the review sections of rokomari.com, a popular online platform for selling books in Bangladesh. The reviews were manually annotated into positive and negative sentiments by experts. We have investigated the performance of four different deep neural network models; LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM, and three transformer models; m-BERT, Bangla-BERT and XLM-R. We have found that XLM-R outperforms the other DNN models as well as the other transformer models achieving the highest weighted f1 score (88.95%) on test data. The dataset is made publicly available at https://github.com/gcsarker/Bangla-Book-Review-Dataset.