smsociety15

Authors: Masoud Makrehchi and Somayyeh Aghababaei

In this study we present how user-generated content in social media can be leveraged in crime trend prediction. We introduce a topic-based model for predicting crime trend. While conventional trend prediction methods are limited to availability of labelled data, our proposed model generates training data without human interventions by labeling with the background knowledge inferred from the domain. A temporal topic identification model was proposed to capture the most “novel” topics with addressing topics evolution. The results of experiments reveal the correlation between inferred topics and crime index changes. While predictability is high in some specific crime types, it could be variant depends on the incidents. In overall, the study provides insight into the impacts of social data in providing predictive indicators for real world problems.