An Application of Sentiment Analysis Techniques to Determine Public Opinion in Social Media

This paper describes a prototype application that gathers textual data from the microblogging platform Twitter and carries out sentiment analysis to determine the polarity and subjectivity in relation to Brexit, the UK´ s exit from the European Union. The design, implementation and testing of the developed prototype will be discussed and an experimental evaluation of the product described. Specifically we provide insight into how events affect public opinion and how sentiment and public mood may be gathered from textual twitter data and propose this as an alternative to opinion polls. Traditional approaches to opinion polling face growing challenges in capturing the public mood. Small sample response and the time it takes to capture swings in public opinion make it difficult to provide accurate data for the political process. With over 500 million daily messages posted worldwide, the social media platform Twitter is an untapped resource of information. Users post short real time messages views and opinions on many topics, often signed with a ‘#hashtag’ to classify and document the subject matter in discussion. In this paper we apply automated sentiment analysis methods to tweets giving a measure of public support or hostility to a topic (‘Brexit’). The data were collected during several periods to determine changes in opinion. Using machine learning techniques we show that changes in opinion were also related to external events. Limitations of the method are that age, location and education are confounding factors where Twitter users over represent a young, urban public. However, the economic advantage of the method over real-time telephone polling are considerable.