Bang-lish sentiment classification using deep learning

Abstract

E-commerce websites and social media platforms have become integral parts of peo ple’s social lives. Through posts, comments, and reviews on social media and online shopping websites, people can share their ideas. Understanding people’s opinions and evaluating input requires the ability to classify sentiment. A variety of deep learning methods have been employed over time to categorize sentiment. Bang-lish, which consists of Bengali words printed in English letters, has gotten very little notice nonetheless. In addition, the majority of Bengali-speaking individuals utilize Bang-lish to post evaluations on e-commerce platforms. Understanding customers’ ideas are crucial for sellers who want to improve their goods. Bang-lish, however, is challenging to understand and evaluate since it lacks a set grammar. Convolutional neural networks (CNN) and gated recurrent units (GRU), two types of deep learning, are combined in this study’s proposed hybrid framework. Additionally, during the data preprocessing stage, we created a spell check algorithm for the most frequently used words and eliminated Bang-lish stop-words. In binary sentiment classification, our suggested model achieved 89% accuracy, 88% precision, 89% recall, and an 89% F1 score.