Real-time recommendation algorithms for crowdsourcing systems

Crowdsourcing has become a promising paradigm for solving tasks that are beyond the capabilities of machines alone via outsourcing tasks to online crowds of people. Both requesters and workers in crowdsourcing systems confront a flood of data coming along with the vast amount of tasks. Fast, on-the-fly recommendation of tasks to workers and workers to requesters is becoming critical for crowdsourcing systems. Traditional recommendation algorithms such as collaborative filtering no longer work satisfactorily because of the unprecedented data flow and the on-the-fly nature of the tasks in crowdsourcing systems. A pressing need for real-time recommendations has emerged in crowdsourcing systems: on the one hand, workers want effective recommendation of the top-k most suitable tasks with regard to their skills and preferences, and on the other hand, requesters want reliable recommendation of the top-k best workers for their tasks in terms of workers' qualifications and accountability. In this article, we propose two real-time recommendation algorithms for crowdsourcing systems: (1) TOP-K-T that computes the top-k most suitable tasks for a given worker and (2) TOP-K-W that computes the top-k best workers to a requester with regard to a given task. Experimental study has shown the efficacy of both algorithms.