11:30am - 12:00pm

Dr. Jelena Tešić

Computer Science, Texas State University

Title: Signed Graph Analysis For Real Data

Abstract: Content bits, health records, service reviews, and expressed opinions do not exist in a vacuum; they are almost always related to one another, depending on the context (time, place, circumstances, communities, etc.) and attitude. The unsigned graph representation does not capture the richness of the relationships in the unstructured data collections. The signed graph extension to date does not provide principled conclusion as the proposed algorithms for signed networks have been evaluated on limited real and synthetic networks that are too similar in topology to support the research progress of signed graph analysis for real data. In this talk, Dr. Tešic introduces the scalable ´ expansion of the balance theory paradigm for scalable signed graph analysis with application to ground truth recovery, continuous vertex clique labeling, edge reliability, and bias identification. The approach extends the mathematical sociology theory for extended signed graph representation and analysis to account for the higher-level context and addresses state-of-the-art algorithmic assumptions, breaking points, and contextual assumptions of models fitted to the signed graphs and introduces new context-free paradigm with scalable implementation to address state-of-art limitations. The work is a joint project with Dr. Rusnak from the Math department.