Crowdsourcing Techniques

Another major goal in the VaVel project is to develop crowdsourcing techniques to complete missing information or complement noisy urban real-time data streams to improve predictability and resiliency.  The techniques we have developed involve active learning for data labeling and task assignment techniques to select the most appropriate human workers and ask the proper questions.  The proposed techniques take into account human crowd reliability and real-time response factors. In D4.2, Section 4 we describe a suite of crowdsourcing techniques developed for the above goal. These techniques are summarized bellow:

• In  D4.2, Section  4.1  and in BK16b we  describe  a  location  detection  technique  where  human  users,  acting as social sensors, can contribute in determining the location, extend and severity of an event.  Our approach is based on the use of particle filters which have significant benefits as they can be used for an unbounded number of variables, the particles can cover well a large space and they can be applied to non-linear models.
• In D4.2, Section 4.2 and in BKG16  we describe our methodology for efficient state detection using sampling techniques, where the approach considers both human factors (e.g., user subjectiveness) as well as the user geographical locations, to maximize both the spatial coverage and the knowledge extracted from the samples when detecting the state of an event.
• In D4.2, Section 4.3 and in BK16a we quantify the problem of privacy exposure of the users when they share their  location-based  data  considering  possible  links  with  heterogeneous  data  sources, and  propose  a  feature-based  entity  resolution  technique  that  can  identify  similar  user identities and react to preserve user privacy.  
• In D4.2, Section 4.4 and in LKKG16a we aim at capturing the influence of the human social sensors and propose techniques for limiting misinformation propagation.

References: