Leveraging Deep Learning in Human Resources: A Systematic Investigation of Graph Neural Networks for Candidate-Job Matching

This dissertation explores the application of Graph Neural Networks (GNNs) to Human Resource Management (HRM), focusing on the challenge of candidate-job matching (CJM). Through theoretical contributions and applied studies, it demonstrates the potential of graph-based approaches to enhance HR analytics and personnel selection. The research addresses key questions on representing HR data using deep learning, translating it into graph structures, identifying effective GNN architectures, and applying them to HR tasks. Novel methods are developed to convert various HR data types into graphs, including Likert-scale questionnaires, candidate profiles, and candidate-job pairs. Theoretical contributions include a topological-based aggregation for GNNs using Generative Topographic Mapping, multiplicative integration as a graph convolution operation, and an efficient algorithm for computing Shapley Interactions in GNNs. These advancements improve GNN performance and interpretability. Applied studies leverage Large Language Models for feature extraction from CVs and job descriptions, combining them with GNNs for predictive modeling. A comprehensive pipeline is developed to process real-world HR data, construct purpose-built graphs for each candidate-job pair, and perform inductive learning for CJM. Results show that graph-based approaches outperform traditional methods in capturing complex relationships in HR data. The research highlights key considerations for applying GNNs to HR, including handling class imbalance, ensuring interpretability, and incorporating domain knowledge. This work bridges cutting-edge machine learning with real-world HR challenges, offering new perspectives for addressing modern recruitment complexities. It emphasizes the importance of keeping humans central in the recruitment process while leveraging AI to augment decision-making, pointing towards a future of more sophisticated, fair, and effective AI-driven recruitment systems.