Aspect Based Sentiment Analysis in Bengali Text using Machine Learning and Deep Learning Algorithms

Sentiment analysis (SA) has been used to monitor social media, help customers, perform market research, and gauge client sentiment to improve businesses’ products and services over time. The conventional approach of SA often focuses on classifying sentiments into three main polarities: positive, negative, and neutral. Aspect based sentiment analysis (ABSA) is an advantageous methodology that facilitates a comprehensive analysis of consumer feedback. This approach not only determines the sentiment polarity of the text but also identifies the specific attributes of the product or service. While substantial research on ABSA has been conducted in English and other widely spoken languages, Bengali-language research on this topic is notably limited. This study explores machine learning and deep learning methodologies for analyzing Bengali text, particularly emphasizing on five different aspects and their corresponding emotions. This research employs several machine learning approaches, such as Decision Tree, Random Forest, XGBoost, Gradient Boosting, and Naive Bayes, to investigate both aspect and sentiment. The Gradient Boost model exhibits the best level of accuracy, reaching 67%, in the task of sentiment detection. On the other hand, the Random Forest model achieves an accuracy of 87.4%, in the task of aspect determination. In addition, several deep learning algorithms, including BiLSTM, BiGRU, and BERT, have been used in this study. Compared to the BiLSTM, and BiGRU models, BERT has the highest sentiment analysis and aspect accuracy, with 73.09% and 88.84%, respectively.