Hybrid Sentiment Analysis on Political Tweets

On social media, the significantly increased number of online users generates a large amount of unstructured text in the form of posts, chats, blogs, and messages. Moreover, information exchange on social media is really convenient for describing opinions that gain popularity when admired by a large number of online users. This popularity may throw back the people’s sentiments towards that organization, place, or person. Twitter generates a large number of texts with political insights that can be extracted for analysing people's opinions and predicting future election trends. The proposed system is implemented with the hybrid political sentiment analysis technique by combining Lexicon-based approach with a machine learning approach. Data from Presidents Obama, Donald Trump, Hillary Clinton, and Joe Biden are collected from Twitter for this system, and Apache Spark is used to analyse this large dataset. To get better performance on the multi-class political sentiment analysis, three supervised learning approaches—Decision Tree, Linear Support Vector Classifier, and Multinomial Naïve Bayes are used. According to the experimental result, the Support Vector Classifier is the best optimal sentiment classifier on a multi-class sentiment analysis system. The overall analysis of the evaluation results shows that the proposed system performs with an accuracy of 94.8% in the environment of Big Data Analytics.