Optimizing Sentiment Classification in Bangla Texts Using Ensemble Models and LSTM Networks

Abstract

Sentiment analysis is a way where huge amount of data can be categorized into different sentiments, emotions, attitudes or opinions. Using sentiment analysis, public views are collected can be given prediction of the class that the views belong to. So, for deriving marketing intelligence sentiment analysis can be performed. It is a vast research field in the NLP sector. Deep learning is part of that machine learning algorithm but it works on in depth similarly like the working behavior of the human brain. The study shows a deep learning approach to sentiment analysis using social media’s Bengali dataset. The dataset is a representation of views of audience of social media’s Bengali contents and is consists of around 2994 Bengali comments extracted from social media named Facebook and YouTube posts and videos. Here Positive, Negative and Neutral classes are used to categorize the Bengali data. Also, Keras tokenizer is used to tokenize the Bengali text and to convert it into integer sequence and Keras model was used to run the deep learning algorithms LSTM and Bi-LSTM. And these deep learning approach such as LSTM and Bi Directional LSTM has been performed on the dataset and Bi Directional LSTM has the highest accuracy of 72.25%.