TDMatcher: A topic-based approach to task-developer matching with predictive intelligence for recommendation

Abstract

Artificial Intelligence is currently gripping the business world, which is the next step on the journey from Big Data to full automation. As crowdsourcing has been widely adopted by more enterprises and developers, the software crowdsourcing platform is able to collect enough data. Therefore, we introduce predictive intelligence to solve complex problems. This provides a bridge between software developers and enterprises: developers look for suitable tasks, whose aim is to gain revenues with respect to their interests and abilities; enterprises look for developers that are able to complete crowdsourcing tasks and/or solve hard problems. One main problem is the prediction challenge, i.e., how to perfectly predict the developers for the software crowdsourcing tasks and make appropriate recommendations. To solve the problem, this paper introduces predictive intelligence and proposes TDMatcher, which can effectively perform task-developer pairs prediction and recommendations for software crowdsourcing. First, we builds a unified model for tasks and developers such that they can be matched in the same domain space. Second, we quantitatively measures the matching degree between tasks and developers. Third, we randomly generates potential matchings between developers and crowdsourcing tasks and then employs an MCMC sampling approach to optimize the whole process. Highly matched task-developer pairs can be achieved in the sampling process. In order to solve the cold-start problem, we constructs a social network for each new developer, which indicates that the developer’s interests/abilities to be modeled We implemented TDMatcher and evaluated it against the state-of-the-art approaches on the real-world dataset. The experimental results clearly demonstrate the superiority of TDMatcher. We measured our proposed TDMatcher through the accuracy, diversity and Harmonic Mean of TDMatcher, and found that: (1) TDMatcher outperforms the state-of-the-arts by 15+% in the prediction accuracy and 30% in diversity; and (2) TDMatcher achieves a balance between accuracy and diversity. We believe that TDMatcher provides crowdsourcing platforms with much more capabilities in finding appropriate developers to complete crowdsourcing tasks or vice versa.