Project Description

Modern sensors or sensor networks provide values for the observation of natural phenomena, but only for in their respective location and for a very limited number of measured parameters. One approach to complement measured sensor values is to use observations by eyewitnesses ("humans as sensors"), which are increasingly spread using social media platforms like Facebook, Twitter, or YouTube and Flickr. Each user of these networks represents a mobile, virtual sensor ("Social sensor").
"Social Sensors" offer advantages over conventional sensors, such as high mobility, high versatility of the parameters that can be recorded, and a very rapid spread of acquired data. In addition, the amount of data provided via social media platforms is extensive. The disadvantage is that these data are often characterized subjectively by the observer, and are of varying quality and quantity. To make acquired data from social media platforms available for use in disaster management, methods and procedures are developed to filter out the amount of information from the appropriate data.
The project aim is to gain data from social media platforms that facilitate, in conjunction with data from conventional sensors, rapid damage assessments, e.g., providing input parameters for damage modeling, and thus permit qualified information for the disaster management. This requires appropriate methods and prototypes which will be developed in cooperation with the CEDIM projects "Crowd sourcing in disaster management", "Rapid Flood Event Analysis" and "Forensic Disaster Loss Analysis".
Metrics are derived and evaluated to determine the relevance of data from social media platforms with respect to specific disaster events. This determines which data is appropriate for further analysis and extraction of measuring parameters. Further, with the help of machine learning classification, methods are developed to rank data according to their content (e.g., information, opinion or emotion) in order to allow quality assessments and evaluations of specific issues and further analysis. In addition, patterns are identified which allow to estimate when which information will be available and in what quality it has. Thus it can be estimated in case of an event, at which point in time reliable information can be expected.