Automating Team Formation Using Machine Learning

Collaborative teams are the primary vehicle for coordinating experts with diverse skills for a particular project in academia, manufacturing, freelancing, and the healthcare sector. Forming a successful team whose members can effectively collaborate and deliver the outcomes within the specified constraints, such as planned budget and timeline, is challenging due to the immense number of candidates with various backgrounds, skills, and personality traits, as well as unknown synergistic balance among them; not all teams with best experts are necessarily successful. Historically, teams have been formed by relying on experience and instinct, resulting in suboptimal team composition due to the incomprehensive knowledge of candidates and hidden cognitive biases, among others. To automate forming optimum teams, current methods perform an exhaustive search over subgraphs of expert collaboration networks. They are not, however, scalable for large networks. In our research group, we propose machine learning models that learn relationships among experts and their social attributes through neural architectures. We aimed at bringing efficiency while maintaining efficacy by employing inherently iterative and online learning procedures in neural architectures. We also aimed at utilizing unsuccessful teams to convey complementary negative signals to neural models. Based on the closed-world assumption, we assume no currently known team of experts for the required skills is to be unsuccessful. Our experiments on two large-scale benchmark datasets, computer science research publications (DBLP) and movies (IMDB), show that neural models that take unsuccessful teams into account are faster and more accurate in forming collaborative teams