Intelligent Career Advisory System: Enhancing Career Matching with BERT and Graph Neural Networks

Traditional career recommendation systems often fail to capture dynamic skill demands, subtle query semantics, and potential ethical and bias issues. In this paper, we propose a hybrid model that integrates BERT for semantic understanding and Graph Neural Networks (GNN) for relational reasoning to construct a dynamically updated career knowledge graph. The graph is built by fusing multi-source data from LinkedIn, O*NET, and synthetic data generated via GPT-4, and a cross-modal attention mechanism is used to fuse text and graph features. Experiments on the CareerQA dataset, consisting of 50,000 annotated queries, demonstrates that our model improves the Fl-score by 18.7% and achieves an NDCG@5 of 0.921. A user study involving 1,200 participants further validates the system's practical effectiveness with an 89.2 % positive feedback rate.