Mining Transaction Streams to Infer Semantic Relations

Research Areas: Human Centric Security,

Principal Investigator: Jennifer Neville

This project focuses on understanding and exploiting information in large-scale, dynamic relational networks. In a growing number of relational domains, the data record temporal sequences of interactions among entities. For example, in social networking sites such as facebook.com, members continuously visit other members’ pages, accessing content and posting comments. These use patterns could be utilized to infer the nature and strength of relationships among members, which may then in turn be exploited to improve personalization efforts, marketing strategies, and system design. Our work will produce the first available data mining tools that can simultaneously exploit both the temporal and relational aspects in streams of transactions. We are developing automated methods to infer high-level semantic relationships (e.g., friend, colleague) among entities from dynamic patterns of low-level transactions (e.g., file transfers, phone calls). We will use these semantic relations to identify and exploit the dependencies among entities, thereby improving the accuracy of predictive models. For example, malfeasance is usually a social phenomenon, communicated and encouraged by the presence of other individuals who also wish to engage in misconduct. Thus, if we know one person is involved in fraudulent activity, then his close contacts have increased likelihood of being engaged in misconduct as well.

Personnel

  • Nesreen Ahmed
  • Pelin Angin

Keywords: relational networks. domains, semantic relations, exploiting information