Social Media Emerging as a Third Eye !! Decoding Users' Sentiment on Government Policy: A Case Study of GST

摘要

Micro blogging sites and other social networking platforms have become the primary means of communication and knowledge sharing with progressing technological trends. People across the globe express their views on products & services, predict share price and present feedback on the policies of the regimes. Everything that is shared on social networks may not be authentic or denote the truth. However, it definitely forms a basis to investigate and comprehend the public sentiments. Public sentiments are capable of affecting the economic landscape via foreign investments and stock markets among having other financial and social impacts. In this paper, we have analyzed public sentiments on the Goods and Services tax, popularly known as GST in India. GST subsumes eight central and nine state taxes thereby integrating the absolute indirect tax framework in the country which paves the way for varied opinions & reactions imperative to analyze a collective sentiment. We used a hybrid approach to do the sentiment analysis which uses a combination of lexicon-based method and supervised machine learning approach to determine public sentiments. We accumulated 163,373 tweets over a span of three weeks from July 4th to 25th, 2017 after GST was implemented in India w.e.f. July 1st, 2017. A spatio-temporal analysis was performed on the collected tweets. In this research, we annotated 22,000 unique tweets with the help of a lexicon-based method and thenceforth applied supervised machine learning techniques with a set of six distinct algorithms to train and predict the polarity on the complete data set. K-fold cross validation technique, for K in range of 3–10, was used to assess the model for an independent data set. Subsequently, it was found that accuracy, precision, recall and F1 score of all the models provided the best results when K approached 10. Resultantly, we observed that SVM and Logistic Regression could predict the polarity of new incoming tweets with an accuracy of 77.6% and 79.31% respectively.