Date of Award
Graph Neural Networks;Information Retrieval;Team Formation
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Establishing a competent team is crucial to the success of a project and is influenced by skill distribution and geographic proximity. A team not only benefits from the shared knowledge amongst the team members derived from geographic closeness but also affects the outcome of the project the team is assigned to perform. A team benefits by sharing resources among each member, collaborating efficiently on a given task, brainstorming on an idea more effectively and saving time and money for both the team members and the organization. This thesis uses a neural-based multi-label classifier after a spatial team formation that uses graph neural networks to transfer information from a heterogeneous collaboration network among experts. Our approach to maximizing the effectiveness of team composition considers the dynamic relationship between members’ shared skill sets and geographic proximity to one another. Specifically, we build a heterogeneous network with the nodes being experts, skills, and places to represent the intricate connections between the specialized knowledge of experts and the regions in which they are present. We use graph neural networks to learn vector representations of skill profiles and geographic proximities using meta paths. Then, we follow that up with a feedforward neural model to recommend a ranked list of experts as a team. Following this pipeline allows us to maximize skill coverage while minimizing geographic dispersion, balancing effective collaboration and efficient communication among team members. We evaluate the accuracy of the recommended teams of experts concerning the requisite abilities and geographical distribution by utilizing classification and information retrieval measures. Our methodology was influential in building skilled and geographically coherent teams, as evidenced by experimental assessments of our suggested method on a real-world dataset of patents and computer science articles compared to baseline methods. We experiment our methodology on uspt and dblp with range of graph and neural architectures across different hyperparameters. The outcomes of this study contribute to the process of team creation by drawing attention to the advantages of using graph neural networks that consider both a person’s skills and their location.
Saxena, Karan, "Geo-location informed Team Formation using GNN" (2023). Electronic Theses and Dissertations. 9200.