Using Importance Flooding to Identify Interesting Networks of Criminal Activity

Abstract

In spite of policy concerns and high costs, the law enforcement community is investing heavily in data sharing initiatives. Cross-jurisdictional criminal justice information (e.g., open warrants and convictions) is important, but different data sets are needed for investigational activities where requirements are not as clear and policy concerns abound. The community needs sharing models that employ obtainable data sets and support real-world investigational tasks. This work presents a methodology for sharing and analyzing investigation-relevant data. Our importance flooding application extracts interesting networks of relationships from large law enforcement data sets using user-controlled investigation heuristics and spreading activation. Our technique implements path-based interestingness rules to help identify promising associations to support creation of investigational link charts. In our experiments, the importance flooding approach outperformed relationship-weight-only models in matching expert-selected associations. This methodology is potentially useful for large cross-jurisdictional data sets and investigations.