Discovering Twitter Sentiments using Correlations among Multiple Terms

Abstract

Twitter micro-blogging website has taken attention of re- searchers, since its birth in 2006. A number of research efforts has been devoted to study different aspects of data published on Twitter. These studies apply several mechanisms and tools to capture, store and analyze Twitter data. The experts and analysts may look into the sentiments of the online users within their comments published on Twitter and under- stand their views towards certain subjects or area of interest. This paper presents a novel lexicon-based sentiment analysis approach to detect polarities of Twitter users through correlations among multiple terms. The proposed approach is the fusion of polarity and topicality for a certain topic. Twitter data based on a certain topical subject are collected in a document, called Twitter Dataset Document (TDD). Association analysis technique has been exploited to build a directed graph of the multiple associated terms present in the TDD. Then, the top ranked terms are used in sentiment analysis with the help of SentiWordNet dictionary. The users’ sentiments are computed based on correlations among multiple terms to facilitate the analysts in understanding the topic oriented expressions of the Twitter users. The results reveal the effectiveness and usefulness of the proposed approach on real text data (i.e., tweets collected in a document).