Preprint Open Access
This paper presents a conceptual model and a proof-of-concept implementation of a novel approach to engage citizens in supervising the analysis of user-generated geographic content (UGGC). For example, UGGC allows insights into the (re-) production of urban spaces, promising new options for pro-active urban planning, citizen participation, and local empowerment. However, the unknown quality and high volume of UGGC require advanced filtering and classification procedures to be able to extract new knowledge and actionable intelligence. The complexity of some tasks and the increasing volume of UGGC often restrict the role of citizens to data providers. We argue that citizen should and can play a decisive role in the parameterization and training of deeper computational analysis of UGGC streams. The challenge is how to present geographical analysis problems to a crowd of human supervisors, and how to elicit responses and feed those back into the workflow. We propose a hybrid processing approach, which maps geographical problems into data mining and machine-learning tasks, presents analysis results to human supervisors, and uses the responses to improve the machine-learning and data mining. For the pilot study, we adapt an approach to find semantically distinct places in UGGC. The human supervisors rate the clustering of potentially similar geo-located photographs from the platform Flickr, and thereby help parametrize both the data mining of geospatial clusters, as well as the classification of similar images based on several ancillary attributes.