Sentiment Analysis of Facebook and YouTube Bengali Comments Using LSTM and Bi-LSTM
Nolak Kapali, Tariquzzaman Tuhin, Anik Pramanik, Md. Sadekur Rahman, Sheak Rashed Haider Noori
Abstract
Recently, sentiment analysis has been performed on various information on social media to derive market intelligence. As we know social media is filled with different content as a result audiences are interacting there making it a huge opportunity to perform sentiment analysis on this information. In terms of Bengali content, many audiences on social media interact with the Bengali language which makes social media a treasure trove to perform sentiment analysis in Bengali Natural Language Processing (NLP) field. Here in this study sentiment analysis has been performed on the audience’s Bengali comments expressing different views towards social media’s Bengali contents. The dataset contains 4000 Bengali comments collected from Facebook and YouTube Bengali Contents. Here positive, negative, and neutral classes are used to categorize the Bengali data, and a tokenizer from the Keras library is used to tokenize the Bengali text. Deep learning algorithm Long Short-Term Memory (LSTM) and BiDirectional Long Short-Term Memory (Bi-LSTM) are performed and Bi-LSTM has the highest accuracy of 97.25% than LSTM.