Visual analysis of volunteered geographic information for interactive situation modeling and real-time event assessment (VA4VGI)

Professor Dr. Thomas Ertl, Dr. Steffen Koch & Dr. Dennis Thom
Universität Stuttgart, Institut für Visualisierung und Interaktive Systeme, Abteilung Graphisch-Interaktive Systeme (GIS), Universitätsstraße 38, 70569 Stuttgart

With the proliferation of social media, many researchers have investigated how the generated content can be systematically utilized to derive semantic sensor information about critical events. Naturally, volunteered geographical information (VGI) - such as geotags or GPS locations provided by mobile phones - plays a pivotal role in making the data useful for situation assessment. However, so far, existing approaches only enable the discovery of indicative information from the data such as the occurrence of critical incidents, relevant situation reports, or specific movement behavior. Consequently, the next step would be to better relate VGI data to the underlying real-world quantities. Examples for this would be deriving the spatial and temporal boundaries of an event, illustrating how this extent evolved spatially over time, estimating a number of affected people in an area, illustrating the current state of evacuation measures, or providing numeric indicators for the severity of a situation.

In this project, we propose a visual analytics approach that enables a more holistic situation assessment based on VGI features of social media. Particularly, we will investigate how the deep integration of visual interactive schemes with automated machine learning tools can be used to create, manage, and adapt scalable ad-hoc regression models during time-critical decision-making. These models can then be used to map the observed large-scale indications to actual real-world phenomena. To enable these capabilities, we will first address the more general problems of data sparsity, location uncertainty, and real-time decay of information. New interactive approaches are required to infer the location of massive amounts of entries based on the content of messages, the network of users, and their own location history. We will adapt interaction schemes like multiple coordinated views and focus+context exploration to the realm of real-time streaming data and we will advance visualization techniques to cope with errors and to convey the uncertainty to analysts in a meaningful way. Based on these building blocks, we will establish an integrated system that allows creating holistic space-time situation representations from the extracted information. By selecting entities like incidents, messages, or movements, we can programmatically compare them to past situations and context information, and normalize the data volumes based on large-scale recorded densities. Depending on data availability, analysts will then be enabled to initiate an iterative process of creating regression models that map social media to real-world quantities, validate them based on their manual assessment, and re-configure or re-train them as needed.