A Data-driven Based Bengali Text Sentiment Analysis for Internet Service Users

The enormous growth of the Internet and ease in social media accessibility has engaged more users in social media platforms where people can share their experiences freely. Sentiment analysis, a widespread contemporary research interest in the field of natural language processing (NLP), is a technique that extracts valuable information from text data and classifies those data into different class labels based on their sentiment. Literature shows that several kinds of research have been conducted on sentiment analysis in different areas, such as helping the telecommunication industry by providing customer reviews for GSM service providers, creating recommender systems for the mobile industry, and so on. However, most of the research has been done on the English language, and under-resourced languages (e.g.) Bengali have been omitted. Hence, this study created a sentiment analysis model utilizing a data-driven approach to identify Bengali sentiments of Internet service users. A large dataset has been created by utilizing the Facebook platform, and several machine learning and deep learning algorithms have been evaluated on the dataset. After a rigorous evaluation, the result shows that LSTM got an accuracy of 87.8%, which proves the suitability of this study. The described procedure is fully automatic and can be leveraged by the Internet service providers to provide insightful suggestions to their customers.