Sentiment analysis of tweets using supervised learning algorithms

Abstract

The proliferation of user-generated content (UGC) on social media platforms has made user opinion tracking a strenuous job. Twitter, being a huge mi-croblogging social network, could be used to accumulate views about politics, trends, and products, etc. Sentiment analysis is a mining technique employed to peruse opinions, emotions, and attitude of people toward any subject. This is conceptualized using digital data (text, video, audio, etc.) or psychological characteristics of humans. This procedure assists in opinion mining without having to read a plethora of tweets manually. The results could be wielded to provide an edge for businesses and governments in rolling out new entities (policies, products, topic, event). Cleaning data is an important step here, which we accomplished using regular expressions and NLTK library in Python. We implemented nine separate algorithms to classify tweets and compare their performance on cleaned data. It was observed that the convolutional neural network produces the most optimal results at 79% accuracy.