A Collaborative Approach to Mobile Crowdsourcing Based on Data Stream Learning

Ralf Bruns, Jeremias Dötterl, Jürgen Dunkel, Sascha Ossowski

Abstract: Mobile crowdsourcing refers to systems where task completion necessarily involves physical movement of crowd workers. Crowdsourced parcel delivery, also called crowdshipping, is a particularly relevant example to this respect. Evidence exists that in such systems tasks are frequently abandoned, indicating that crowd workers accept tasks that they misjudge and later prefer to discontinue. In this paper we evaluate as to how far on-the-fly task transfers between crowdworkers can alleviate this problem in a cooperative setting. Its contribution to this respect is twofold. Firstly, it analyses different data stream learning approaches for service quality prediction in mobile crowdshipping. Secondly, it embeds this prediction model into a collaborative agent-based crowdshipping approach where task transfer decisions are taken in a peer-to-peer fashion with limited overhead.