MSc Thesis Proposal by Puja Sharma

Abstract: Team formation is critical for achieving objectives across various domains, requiring a blend of experts with complementary skills. Traditional methods, relying on graphs or neural networks, often overlook individual skill levels, leading to suboptimal teams. In our research, we introduce a novel framework leveraging Graph Neural Networks (GNNs) to predict individual skill strengths. This enhances team formation accuracy by considering both skill requirements and individual expertise, resulting in more precise team compositions.

Our approach introduces a new method for team recommendation by considering the individual skill strengths of each expert. We break down our approach into three main parts: Expert modelling, Skill modelling, and Skill strength prediction. In the Expert modelling part, we analyze two types of graphs commonly seen in team recommender systems: the Expert collaboration Network graph and the Expert-Skill graph. This allows us to understand experts from different angles. We use two types of aggregations: Skill aggregation, which looks at how experts and skills interact in the Expert-Skill graph, and Expert collaboration Network Aggregation, which examines relationships between experts in the Network graph. By combining information from both graphs, we can better understand each expert's abilities. The Skill modelling component focuses on understanding the skills themselves. To do this, we introduce expert aggregation, which combines an expert's skill strength with their interactions in the expert-skill graph. Finally, we use prediction to learn model parameters by combining insights from both expert and skill modelling components.

In our proposed approach, the skill embedding, augmented with the corresponding skill strength, serves as input to the Variational Bayesian Neural Networks (VBNN) for team recommendation of experts. This approach improves team recommendation by considering not just the skills required but also the individual skill strengths of experts, leading to more effective team compositions.