Detecting Anomalies in Time-Varying Media Crime News Using Tensor Decomposition

Abstract

Nowadays, the mass media surround us in many forms. Newspapers, radio and TV reports about many topics, including the crime committed in a region. Indirectly, the media provide statistics about crime incidents, and policymakers could focus their attention on the unusual number of crime news (c.f., regular events) for evaluating and proposing new public policies. In the present work, the Tensor decomposition is used to detect an unusual amount of crime news. To achieve this goal, two rejection criterion techniques were compared. Also, several image binarization techniques were used to validate our proposal. Our result can be used to detect an unusual amount of crime news as a proxy of unusual crime activity.