Community Detection and Influence Maximization in Online Social Networks

Kanna Gharib Al-Falahi


The detecting and clustering of data and users into communities on the social web are important and complex issues in order to develop smart marketing models in changing and evolving social ecosystems. These marketing models are created by individual decision to purchase a product and are influenced by friends and acquaintances. This leads to novel marketing models, which view users as members of online social network communities, rather than the traditional view of marketing to individuals.

This thesis starts by examining models that detect communities in online social networks. Then an enhanced approach to detect community which clusters similar nodes together is suggested. Social relationships play an important role in determining user behavior. For example, a user might purchase a product that his/her friend recently bought. Such a phenomenon is called social influence and is used to study how far the action of one user can affect the behaviors of others. Then an original metric used to compute the influential power of social network users based on logs of common actions in order to infer a probabilistic influence propagation model. Finally, a combined community detection algorithm and suggested influence propagation approach reveals a new influence maximization model by identifying and using the most influential users within their communities. In doing so, we employed a fuzzy logic based technique to determine the key users who drive this influence in their communities and diffuse a certain behavior. This original approach contrasts with previous influence propagation models, which did not use similarity opportunities among members of communities to maximize influence propagation. The performance results show that the model activates a higher number of overall nodes in contemporary social networks, starting from a smaller set of key users, as compared to existing landmark approaches which influence fewer nodes, yet employ a larger set of key users