Accident monitoring framework based on online social network sensing

Social sensing is increasingly becoming a viable addition to the urban monitoring toolkit for practitioners and decision-makers. It seems to be more flexible and cost-effective as compared to dedicated monitoring systems based on instrumental sensors and surveillance cameras. However, benefitting from these advantages requires deploying fine-tuned approaches to work with rich and voluminous data generated by volunteers or users of social network sites. In this paper, we consider the issue of processing unstructured, distorted and fragmentary textual data produced by the members of the community aimed at sharing experiences and observations of accidents and unusual experiences with the urban environment. We propose a solution that allows aggregating unstructured and noisy user reports and extracting valuable information from them. This solution is based on recurrent neural networks and represents a framework handling the whole of the process starting from data collecting to classifying extracted information.