An automated approach to mining and visual analytics of spatiotemporal context from online media articles

Abstract

Traditionally spatio-temporally referenced event data was made available to geospatial applications through structured data sources, including remote sensing, in-situ and ex-situ sensor observations. More recently, with a growing appreciation of social media, web based news media and location based services, it is an increasing trend that geo spatio-temporal context is being extracted from unstructured text or video data sources. Analysts, on observation of a spatio-temporal phenomenon from these data sources, need to understand, timeously, the event that is happening; its location and temporal existence, as well as finding other related events, in order to successfully characterise the event. A holistic approach involves finding the relevant information to the phenomena of interest and presenting it to the analyst in a way that can effectively answer the what, where, when and why of a spatio-temporal event. This paper presents a data mining based approach to automated extraction and classification of spatiotemporal context from online media publications, and a visual analytics method for providing insights from unstructured web based media documents. The results of the automated processing chain, which includes extraction and classification of text data, show that the process can be automated successfully once significantly large data has been accumulated.