Recommending Teammates with Complementary Skills via Matrix and Tensor Decompositions

초록

?We study the problem of recommending wellcomplemented team members via matrix- and tensor-based techniques. Given a pool of experts with different skills, the problem requires recommending a specific number of experts to one or more designated team member/s to cover a given project. Existing research suffer from neglecting the complementarity concept, the most critical factor that affects the collaboration between recommended teammates and other team members. In this paper, we fill this gap by devising a computational framework to propose experts who most likely have the highest complementarity with the designated team members. To do so, we propose a measure to calculate the complementarity values between adjacent experts according to their skills. Using different matrix and tensor factorization approaches for complementaritybased relations, we bring the graph’s inherent structure into consideration for recommending teammates. Through several experiments on link prediction tasks, we evaluate the performance of both matrix- and tensor-based approaches.