Detection of Topic-Based Opinion Leaders in Microblogging Environments
Recently, the world has become a huge virtual and social environment where people spend a great deal of time expressing their thoughts, feelings and opinions. This virtual socialization seems to bring us to a new era where valuable virtual information is being accumulated. Social networking applications, especially microblogging sites, are the leading actors of this data accumulation. Their free format characteristics lead different kinds of opinions to emerge, interact and broadcast rapidly. In this perspective, detecting opinion leaders, that is people whose opinions are followed, accepted, or taken into consideration, has become crucial in various domains such as marketing, advertisement, and politics. In this research, we focused on identifying topic-specific opinion leaders in Twitter. We extracted four different relationship types, namely retweet, mention, reply, and follow, between Twitter users who were interested in specific topics. Then we formed weighted and directed topic-based social graphs by combining these relationships to compute the edge weights. In order to detect opinion leaders, we applied social network analysis methods including PageRank, betweenness and closeness centrality metrics. We used sentiment analysis methods to evaluate the detected opinion leader candidates. The overall topic-based sentiment and opinion change of the community guided us whether the candidates are the real influencers in a predefined topic or not.