Online Social Networks (OSN) such as Facebook, Twitter, as well as a plethora of online datasets, such as Wikipedia and US government?s DATA.GOV initiative are becoming the repositories of knowledge and discussion for the future. These OSNs and online datasets all share the common feature that they can be thought of as ?Online Information Graphs? (OIG), in the sense that the information embedded in them has a natural graph structure, such as Facebook?s graph of ?friends?. This project considers the question of detection and classification of real-world events using these Online Information Graphs from a communication theoretic viewpoint, by using the graph to specify a Bayes prior. As more and more data is made freely available on the internet, utilizing it to extract information becomes important for society, which is the main goal of the project. For example, Open Government data can improve governance, if inferences applicable to society can be made from it.
The questions about OIG that are being investigated are; defining and learning suitable models for the Bayes prior of the OIG, defining various detection and classification problems on OIGs, designing detection algorithms for large OIGs, analyzing detector performance, and engineering the information graph to improve detection. The research is partly experimental (e.g., exploring real-world datasets) and partly theoretical (e.g., analyzing detector performance), and so, intellectual contributions will be made in both categories. Algorithms are being designed to operate on these models, for graph structures typical in OSN-type datasets. The research uses analytical techniques from communication theory, adapted to probabilistic graphical models, to predict error performance and to engineer the networks.
