MSc Thesis Proposal Announcement of Sagar Kaw: "Skill-based Experts Ranking using Graph Neural Networks"

 

Date: Thursday October 13, 2022 

Time: 1:00PM – 2:00PM 

 

Given a task which requires a set of skills for its completion, the objective of the team formation (TF) problem is to form a team of experts that cover the required skills. There are two recognized fields of interest in this problem: Operation Research (OR) and Data Mining (DM). In OR - researchers tried to find the best match for the task by considering each expert’s attributes individually, whereas, in DM - a social network (SN) of connections between the experts is considered in addition to their attributes to find the best match for the task. 

 

In this research, we focus on the problem of TF in the DM domain. This problem has been proved as NP-Hard. Prior works in this domain solved TF as an optimization problem using computationally expensive graph search methods such as minimum spanning tree etc. Recently researchers have used deep learning (DL) approaches for TF using two-step neural network architectures - i) metapath2vec - a random walks-based SN representation learning ii) variational bayesian neural network - uses the representation from metapath2vec for ranking experts in the SN. This approach faces three limitations: i) The computation time of metapath2vec is dependent on the size of the SN.  ii) It is transductive in nature iii) it doesn’t provide a means to incorporate SN node’s features. Therefore, we propose to use Graph Neural Network (GNN), which is the state-of-the-art DL technique for graph representation learning to overcome the above limitations of the existing DL-based approaches. 

 

Keywords: Team formation Problem, Data Mining, Graph Neural Networks

 

Internal Reader: Dr. Pooya Moradian Zadeh         

External Reader: Dr. Bharat Maheshwari              

Advisor: Dr. Ziad Kobti 

Co-Advisor: Dr. Kalyani Selvarajah