Sentiment Analysis Using Deep Learning: A Comparative Study
DMilind Soam, Sanjeev Thakur · 2022 Second International Conference on Computer Science, Engineering and Applications (ICCSEA) · 2022
Sentiment Analysis is a technique associated with Natural Language Processing in understanding the sentiments of the people. In today’s environment where shopping, watching movies, expressing thoughts or emotions is done via social platforms. The social platforms have become a stage where people can express their sentiments freely. Such behaviour has made it important for companies to provide the best of the platforms for their customers while using their services based on the polarity of their sentiments. In other cases where the manner of the sentiments is to be understood for security reasons or regarding any social issues, sentiment analysis has been the most important tool. In this work we have studied NLP, different deep learning techniques such as LSTM, BiLSTM and CNN along with Word Embeddings and their types: Glove and Word2Vec and performed a comparative study. We found improvement in BiLSTM and LSTM models using Glove(50d) word embeddings. [1]The accuracy received using LSTM and BiLSTM are: 90.36% using batch size 100 and 91.68% using batch size 128. We also tested the best model which was BiLSTM with batch size of 128 on an unlabelled testing data and also projected the results of it. The results were very impressive and promising and at the end the word clouds were created for all the reviews, positive and negative reviews respectively in the unlabelled dataset.