Suspicious Local Event Detection in Social Media and Remote Sensing: Towards a Geosocial Dataset Construction

Abstract

Remote sensing is a powerful technology for earth observation. However, the spatial, spectral, and temporal resolution of the imagery are imposing various limits. Lately, with the rise of the internet and smart mobile devices, social media with location-based information has been rapidly emerging. These circumstances led to the prevailing of new scenarios where fine-grained details of social bookmarking websites are enhanced with the wide coverage of satellites. Social media and satellites are both valuable sources of data. An event-driven data, designating either normal common events or unusual suspicious ones that may threaten human lives or damage the infrastructure. In this paper, we provide an insight into the present state of knowledge to better address the task of local suspicious event detection and linking social media with satellite imagery. Also, to track suspicious local events, we treated the detection problem as a retrospective problem by training different classifiers on the crisisLexT26 dataset. Furthermore, we introduced how to use the available geo-locations in the dataset to construct a geo-social dataset by linking it with remote sensing and retrieving satellite imagery before and after the event occurrence.