Twitter Sentiment Based Mining for Decision Making using Text Classifiers with Learning by Induction

The amount of data residing in social media currently untapped is certainly limitless as millions of people are constantly posting a message or the other to public forums on the internet. Twitter being one of the largest social media networks with over 336 million monthly active users has proven to be a fertile ground for harvesting opinion from multiple people. This work explores how opinion can be extracted from tweets to discover people's view concerning a certain subject matter. It focuses mainly on overcoming the limitation of the current approach to social media sentiment based mining for decision making which is that opinions derived from multiple sources are limited to available connections on the social media platforms and lack of improved accuracy of mined opinions. In order to achieve this, the proposed framework provides a platform to mine opinions from more than the available friends and connections on the social media platform and in addition, improve the quality of the opinion mined by implementing supervised learning algorithms with learning by induction in Twitter data analysis. In this research, three different supervised machine learning algorithms were applied to a dataset curated by graduate students at Stanford in order to accurately classify tweets into either positive or negative sentiment based on its content. It was discovered that Maximum Entropy had the highest accuracy of 83.5% among the three algorithms. The research has provided a web application which would enable users such as CEOs, Market Analysts, and random users make quality decision based on others' opinions.