COSINT: Mining Reasons for Sentiment Variation on Twitter using Cosine Similarity Measurement

Abstract

An advanced domain has evolved in the field of research over the past decade, called Sentiment Analysis on Social Media. Twitter has made huge impact with more than 500 million Tweets each day. People share their opin- ion in the form of Tweets on many topics. In this paper, we employ Foreground and Background LDA (FB-LDA) and Reason Candidate and Background LDA (RCB-LDA) model to extract the reasons for sentiment variation. Emerging topics or Foreground topics within the sentiment variation period are highly related to the reasons for sentiment variation, whereas Background topics are discussed from long time and does not add to the sentiment variation. FB-LDA model filter out Background topics from the Foreground tweet set and extract the required Foreground topics that contribute for the reason for sentiment variation. RCB-LDA model finds more relevant tweets of the Foreground topic that are extracted in FB- LDA model and rank them to get Reason Candidates. To extract Reason more precisely from Reason Candidates, in this paper we propose COsine SImilarity MesuremeNT (COSINT) using Latent Semantic Analysis methods. This methods mine specific reason for sentiment variation.