MULTIMEDIA SOCIAL NETWORKS

Nowadays, On-Line Social Networks represent an interactive platform to share -- and very often interact with -- heterogeneous content for different purposes (e.g to comment events and facts, express and share personal opinions on specific topics, and so on), allowing millions of individuals to create on-line profiles and communicate personal information. In this dissertation, we define a novel data model for Multimedia Social Networks (MSNs), i.e. social networks that combine information on users -- belonging to one or more social communities -- with the multimedia content that is generated and used within the related environments. The proposed data model, inspired by hypergraph-based approaches, allows to represent in a simple way all the different kinds of relationships that are typical of these environments (among multimedia contents, among users and multimedia content and among users themselves) and to enable several kinds of analytics and applications. Exploiting the feature of MSN model, the following two main challenging problems have been addressed: the Influence Maximization and the Community Detection. Regarding the first problem, a novel influence diffusion model has been proposed that, learning recurrent user behaviors from past logs, estimates the probability that a given user can influence the other ones, basically exploiting user to content actions. On the top of this model, several algorithms (based on game theory, epidemiological etc.) have been developed to address the Influence Maximization problem. Concerning the second challenge, we propose an algorithm that leverages both user interactions and multimedia content in terms of high and low-level features for identifying communities in heterogeneous network. Finally, experimental analysis have been made on a real Multimedia Social Network ("Flickr") for evaluating both the feasibility of the model and the effectiveness of the proposed approaches for Influence Maximization and community detection.