Social network analysis : detection of influencers in fashion topics on Twitter

Abstract

Online social networks have facilitated the interaction and topic discussion. Some of this content became a rich and important source of information and also strategical for companies. One of the most popular of such websites is Twitter. Nowadays, the use of digital influencers became a new strategy in the development and in the management of marketing campaigns for leading brands and companies. Fashion industry usually targets them to market products or to diffuse messages. In consequence, the identification of these persons became a central issue for marketers. In this dissertation, I propose a state of research in centrality measures, developed in social network analysis, in order to identify those influencers. Interpretability, robustness and accuracy, current applications and related work on Twitter will be discussed in order to select and understand these concepts. Moreover, I propose a new technique to collect Twitter data with a friendship graph and with a given topic. I perform this research on fashion industry which has not been treated yet in the literature, and then, I use centrality measures to identify the most influential users. The experimental evaluation shows that the presence of reciprocity can be explained by phenomenon of homophily. This finding valids the extraction process to create a sample composed of users interested and influent in fashion topics. The application of centrality measures on the sample provides a relevant ranking of influencers that can be used in a marketing campaign. Keywords: Twitter, Centrality measures, Social network analysis, Degree centrality, Closeness centrality, Betweenness centrality, Eigenvector centrality, Influencer, Network typology, Digital influencer marketing