Sentiment classification on Bengali food and restaurant reviews

Abstract

Sentiment analysis, a critical facet of Natural Language Processing (NLP), plays a pivotal role in decoding human emotions conveyed through text. Despite extensive research in sentiment analysis for widely spoken languages, there is a notable gap in understanding its application to languages with fewer computational resources, such as Bangla. This study bridges this gap by employing deep learning techniques to analyze sentiments in Bangla texts. Our objective is to unravel text encoded in Bangla expressions using a diverse set of machine learning and deep learning models, including Random Forest Classifier, K-Nearest Neighbors (KNN), Kernel-Support Vector Machine (SVM), Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), Convolutional Neural Networks (CNNs), Gated Recurrent Units (GRUs), and BERT-base and RoBERTA and a custom-made model. Among these, our findings reveal that the 1D CNN model achieved the highest accuracy, outperforming all other models with an accuracy of 87.3%. These models underwent training with a custom dataset from various online resources and authentic testimonials. Focusing specifically on food and restaurant reviews in Bangla, we recognize the substantial role customer sentiments play in shaping the food industry. Additionally, a custom model was developed to enhance sentiment analysis in Bangla further. Beyond technical aspects, our research contributes to the understanding of Bangla language sentiment expression nuances. We anticipate that our findings will enrich the field of sentiment analysis, offering insights into linguistic diversity in NLP and inspiring advancements for languages underrepresented in computational research.