Survey of Multimedia Information Mining Techniques Based on Deep Learning in the Context of Crime Domain

The rapid expansion of multimedia data has made multimedia data mining a crucial area of research in the field of crime. In the realm of crime, the significance of multimedia data mining lies in its ability to reveal hidden patterns, correlations, and knowledge within criminal activities, providing powerful support for crime investigation and prevention. For instance, through similarity search, multimedia data relevant to criminal suspects can be swiftly identified. By identifying associations and connections between criminal organizations, criminal networks can be discovered. Through entity resolution and classification, individuals, locations, and items involved in criminal events can be automatically identified. The automatic feature learning and large-scale data processing capabilities of deep learning make it closely intertwined with data mining. Therefore, this paper summarizes the applications of deep learning-based multimedia data mining techniques in the field of crime since 2018, along with the compilation of relevant crime datasets and evaluation metrics. Additionally, it outlines the challenges present in this domain and provides an outlook on future developments.