Ching Hei Wong Undergraduate Dissertation 2016/17

Sentiment Analysis in Twitter

Supervised by A.Vlachos

Abstract

Sentiment Analysis has been an important area of study among the researchers for understanding the trends and opinions of people across the social media. In the last decade, social media like Twitter has grown significantly along with the increased use of the Internet. Methods like sentiment classification of Tweets were adapted to study the sentiment classes in Twitter. However, researchers often focus on prediction of the overall distribution of sentiment classes on certain topics, as known as sentiment quantification, rather than classifying the sentiment of individual Tweets. The aim of this study is to conduct sentiment quantification based on the Tweet data in the well-known Semantic Evaluation 2016 project (SemEval, 2016) and explore the potential of sentiment quantification. The quantification is conducted as binary (2-point) and ordinal (5-point). In this study, a prediction of sentiment on unlabelled Tweets on a certain topic has been done successfully and the sentiment quantification was conducted based on the classification methods. The relative accuracy and error analysis of different algorithms used is documented to justify the reliability of the predictions of the labels on Tweets.