Natural Language Processing and e-Government: Extracting Reusable Crime Report Information

Abstract

Crime reporting needs to be possible 24/7. Although 911 and tip-lines are the most publicized reporting mechanisms, several other options exist, ranging from in-person reporting to online submissions. Internet-based crime reporting systems allow victims and witnesses of crime to report incidents to police 24/7 from any location. However, these existing e-mail and text-based systems provide little support for witnesses' memory recall leading to reports with less information and lower accuracy. These systems also do not facilitate reuse and integration of the reported information with other information systems. We are developing an anonymous Online Crime Reporting System that is designed to extract relevant crime information from witness' narratives and to ask additional questions based on that information. We leverage natural language processing and investigative interviewing techniques to support memory recall and map the information directly to a database to support information reuse. We report on the evaluation of the Suspect Description Module (SDM) of the system. Our interface captures 70% (recall) of information from witness narratives with 100% precision. Additional modules will follow the design and development methods used with this module.