Date of Award

7-8-2024

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Supervisor

Ziad Kobti

Supervisor

Kalyani Selvarajah

Abstract

In recent years, Graph Neural Networks (GNNs) have been found highly effective for learning from graph data in various applications. Real-world graph data often showcase a variety of graph structures and heterogeneous nodes and edges. This thesis considers a collaboration network of experts who worked in the past to achieve a common goal and recommend team of experts with complementary skills who can work together for specific task in the future. The advantages of GNNs provide great potential to advance team recommendation since data in collaboration networks can be represented as expert-expert graphs and expert-skill graphs, and learning latent factors of experts and skills is the key to successful team recommendation. However, building team recommender systems based on GNNs faces challenges: 1) the expert-skill graph encodes both interactions and their associated skill strength; 2) experts are involved in two graphs, the expert-expert collaboration graph and the expert-skill graph. In this thesis, we propose a graph neural network with a Variational Bayesian Neural Networks (VBNN) framework, to address the challenges in team recommendations. In particular, we jointly capture skill aggregation in the expert-skill graph and expert collaboration Network Aggregation in the expert-expert graph and propose the framework to predict skill strengths. The skill strengths are utilized in VBNN (variational Bayesian neural network) to determine the successful team of experts. Extensive experiments on the DBLP dataset demonstrate the effectiveness of our proposed framework.

Download

Share

COinS