Detection and analysis of overlapping community structures for modelling and prediction in complex networks

Many real-world systems are known as complex networks that can be modeled by networks of interacting agents. In complex networks, vertices can be a member of densely connected components named community structures, which sparsely connect to the rest of the net. Community structures can be overlapping, such that vertices are the member of more than one community. We can see examples of overlapping communities in real life, for instance, a scientist is not only active in scientific communities but also may belong to communities of relatives, family members, colleagues, a sports club, etc. We can also observe overlapping communities in other types of networks, e.g., biological and online social networks. Overlapping members often help communities to scale up and to share their knowledge, e.g., diffusion of information in learning environments. In this dissertation, we propose overlapping community detection algorithms that use properties such as degree mixing and information diffusion. We also build prediction models for the evolution of overlapping communities. Besides, we show the importance of overlapping community structures in the prediction of mixing patterns in networks. For this purpose, dynamics of overlapping community structures are used to propose ranking and recommender models. Our proposed algorithms in many cases outperform state-of-the-art techniques. Moreover, the algorithms have been made widely available in our WebOCD software framework. Different applications can use the algorithms proposed in this thesis. Recommender systems can use overlapping community structure dynamics to recommend items to users in overloaded information spaces. Additionally, overlapping community structures can contribute to recommend overlapping members to experts that this helps to increase the information flow and to scale up communities. Algorithms, WebOCD framework, and applications have been validated in several settings, including large-scale informal Learning networks in the European Learning Layers project