Beyond behavioural signals: personality-aware graph neural networks for holistic person-job fit

The job recommendation landscape is increasingly complex due to the inherent heterogeneity of user profiles and job descriptions, alongside the subtle influence of psychological compatibility in workplace success. Traditional collaborative filtering and content-based approaches often fail to capture the deep structural relationships or the nuanced behavioural traits required for optimal person-job fit. This study proposes TransGNN, a hybrid framework that synergizes Graph Neural Networks (GNNs) with Transformer-based self-attention to resolve the inherent search friction in modern recruitment. Unlike traditional models that rely on surface-level keyword matching, TransGNN constructs a personality-aware latent space by anchoring user profiles to 16 granular work styles derived from O*NET occupational data. To bridge the semantic gap between resumes and graph topology, Self-Supervised Learning (SSL) strategy was employed with InfoNCE loss, aligning highdimensional textual features with structural embeddings. Evaluation demonstrates that while collaborative signals remain a robust predictor, the integration of psychological traits through Personality-driven Expert Routing and Adaptive Gating provides a critical advantage in representing complex candidate personas, achieving significant performance gains over standard neural baselines in dense scenarios and offering a more holistic approach to person-job alignment.