Twitter Sentiment Classification with Deep Learning LSTM for Airline Tweets

As a result of increase in internet usage, there is a massive amount of information available to web users, as well as a massive amount of new information being created daily. To facilitate internet pick-up, trading ideas, and disseminating assessments, the internet has evolved into a stage of large volumes of data. Facebook, and Twitter generate a lot of data every day. As a result, text handling is crucial in making decisions. Sentiment analysis has surfaced as a method for analyzing Twitter data. In this paper, we collected a Kaggle dataset with airline tweets. It contains three variants of tweets: neutral, positive, negative. First, we used NLP methods to clean the text data. Later, we applied RNN, LSTM, stacked LSTM, bidirectional LSTM, and GRU techniques for classifying tweets in three different ways: positive vs negative sentiment analysis, neutral vs positive sentiment analysis, and neutral vs negative sentiment analysis. We achieved an accuracy of 93% for the classification of positive and negative airline reviews. We achieved accuracy values of 84.5%,83.8% for neutral vs positive and neutral vs negative tweets. The results show that, the proposed RNN/LSTM/GRU model performed well for sentiment classification. Keywords-Sentiment analysis, Tweets, RNN, LSTM, GRU.