Collaborative Labor Market is the underlying paradigm for a large number of popular web services. By introducing the power of online crowd, many far-reaching real-world applications, such as crowdsourced question answering and ride-sharing, are now effectively conducted at low cost. Despite the current flourish of collaborative labor market, the Quality of Service (QoS) and Throughput of Service (ToS) remain its central issues. Motivated by such a point, we target on addressing the above issues from the perspective of work-force recommendation. Particularly, by allocating appropriate work-force for people’s demands in collaborative labor market, quality service can be timely generated at high throughput rate.

In our work, work-force recommendation strategies are studied in depth...[ Read more ]

Collaborative Labor Market is the underlying paradigm for a large number of popular web services. By introducing the power of online crowd, many far-reaching real-world applications, such as crowdsourced question answering and ride-sharing, are now effectively conducted at low cost. Despite the current flourish of collaborative labor market, the Quality of Service (QoS) and Throughput of Service (ToS) remain its central issues. Motivated by such a point, we target on addressing the above issues from the perspective of work-force recommendation. Particularly, by allocating appropriate work-force for people’s demands in collaborative labor market, quality service can be timely generated at high throughput rate.

In our work, work-force recommendation strategies are studied in depth for the following two application scenarios.

First of all, we study the application of crowdsourced Q&A services, where workers need to be recommended for people’s questions of interest. Given such a problem, we come up with the triple-factor aware approach, which characterizes workers with their activeness, preference and expertise. On top of the above factors, worker recommendation is judiciously generated to maximize the timely acquisition of high-quality answer. According to experimental studies on the Stack Overflow dataset, the exploitation of triple-factor significantly improves the recommendation effectiveness in terms of answer quality and throughput.

Secondly, we work on the application of context-aware academic collaborator recommendation, where new potential collaborators are suggested w.r.t. people’s interested research topics. Inspired by the success of representative learning on graph, we come up with the collaborative entity embedding network, which deeply excavates the researchers’ relationship in academia and research topics’ semantic meaning. To further improve the performance in finding new collaborators, we propose a probabilistic graphical model to take advantage of researchers’ inherent activeness and conservativeness. With experimental studies on the Aminer dataset, it is verified that the effectiveness of finding academic collaborators is greatly enhanced with our proposed mechanisms.