MSc Thesis Defense by: Zahra Kheiryashkuh

In the realm of collaborative work, the optimal assembly of teams is paramount for success. The challenge of constructing cohesive teams lies in selecting from a vast array of potential candidates, each with distinct skills, experiences, and personal attributes. Team recommender systems focus on identifying an optimal assembly of experts who, together, fulfill the necessary skill set to achieve a shared objective. Recently, researchers have started to examine this problem through neural architectures that recommend the team of experts by learning a relationship between the skills and experts space. Yet, such models have been shown to suffer from popularity bias, which refers to the tendency of the systems to recommend disproportionately more popular teams. To improve the performance of team recommendation, we leverage the dual transfer method which transfers knowledge from head instances to tail instances in both model-level and instance-level. The knowledge transfer at the model level establishes a meta-mapping from few-shot to many-shot models. This process implicitly augments data at the model level, enhancing the representation learning for less popular teams. At the instance level, the transfer bridges popular and not popular instances by leveraging curriculum learning. This guarantees a seamless transfer of meta-mapping from frequently head teams to the tail ones. Our evaluation criteria is that we expect to improve team recommendation quality, particularly for teams that are in the tail of the distribution. We demonstrate how the proposed platform overcomes the popularity bias challenge in the existing approaches and compare it against the state-of-the-art approaches in terms of effectiveness using the DBLP dataset.

 

Internal Reader: Dr. Samet         

External Reader: Dr. Hassanazdeh          

Co-Advisors: Dr. Kobti and Dr. Selvarajah

Chair: Dr. Khan