A novel cognitive computing technique using convolutional networks for automating the criminal investigation process in policing

Criminal Investigation (CI) plays an important role in policing, where police use various traditional techniques to investigate criminal activities such as robbery and assault. However, the techniques should hybrid with the use of artificial intelligence to analyze and determine different crime types for taking actions in real-time. In contrast with the manual process of investigating a large amount of data collected related to a criminal investigation. In this paper, we present a novel Cognitive Computing enabled Convolution Neural Networks (CC-CNN) approach for identifying crime types, such as robbery and assault, collected from unstructured textual data. We develop learning algorithms and provide a cognitive assistant to assist a police investigator in easily understanding crime types. We train and validate the CC-CNN technique on two datasets including handcrafted text-crime dataset and sentiment polarity dataset of negative and positive reviews. The experimental results show that our approach performs at a high level in terms of accuracy, error rate and time processing using both datasets.