Online social networks, such as Facebook, Twitter and Foursquare, have become increasingly popular in recent years. These online social networks contain abundant information about the users and their activities. Nowadays, to enjoy more social network services, people are getting involved in multiple social networks simultaneously. However, the accounts of the same user in different social networks are mostly isolated without any connection or correspondence to each other. This project has the potential to make fundamental, disruptive advances in fusion of heterogeneous networks for synergistic knowledge discovery. The success of this project will dramatically extend and change the current social network studies in data mining. In addition to social network analysis, this work can also be beneficial to scientific research such as life sciences on biological networks. The analytic tools developed and data collected will be made available to the public for free download.
The team will investigate the principles, methodologies and algorithms for the synergistic knowledge discovery across multiple partially aligned social networks, and evaluate the corresponding benefits. They plan to address the challenge on effective transfer of relevant knowledge across partially aligned networks, which will depend upon not only the relatedness of the different networks, but also the target application, e.g., link prediction vs clustering vs information diffusion. A general methodology will be developed, which will be shown to work for a diverse set of applications, while the specific parameter settings can be learned for each application using some training data. The problems studied include (1) Partial Network Alignment, (2) Integrated Anchor and Social Link Prediction, (3) Mutual Clustering, and (4) Cross-Networks Influence Maximization. This proposal will address these four major research problems systematically based on a unified concept: integrated anchor and social meta paths (including both intra- network and inter-network meta paths) for relationship exploration.
