Using dynamic community detection to identify trends in user-generated content

Rémy Cazabet, Hideaki Takeda, Masahiro Hamasaki, Frédéric Amblard

Abstract: In this paper, we present a new solution for trend detection in user-generated content, and more particularly Web 2.0 social networks. Whereas some propositions have been published in this domain recently, we have chosen a new approach based on network analysis. We first create an evolving network of terms, which is an abstraction of the complete network, and then run a dynamic community detection algorithm on this evolving network. In order to be able to detect not only short, bursting events, but also more persistent topics, we test our solution on a social network for which we have information about all published contents for a period of more than 2 years: the Japanese network Nico Nico Douga. After presenting our solution in detail, we present the results on this dataset, notably a statistical analysis of communities’ sizes and durations, examples of detected communities, and a typology of the different kinds of trends detected. Finally, we discuss the advantages and disadvantages of this method, as well as its possible applications.