This study presents a novel Hybrid Attention-based Temporal Graph Convolutional Network (HAT-GCN) for forecasting labour market demand and supply under the context of big data analytics. The model addresses critical challenges such as workforce mismatch, rapid technological changes, and fluctuating industrial demands that limit the accuracy of traditional forecasting methods. By integrating temporal learning and graph-based spatial modeling, HAT-GCN captures complex interdependencies among industries, regions, and skill categories over time. The framework utilizes multi-source data, including job advertisements, social media discussions, scholarly publications, and economic indicators, enabling a holistic understanding of labour dynamics. Unstructured textual data are processed through Natural Language Processing (NLP) to extract emerging skills and sectoral trends, ensuring the system’s adaptability to evolving market signals. Empirical results using national labour datasets reveal that HAT-GCN achieves superior predictive accuracy with R2 ≈ 0.90, precision of 0.92, and recall of 0.90, outperforming traditional LSTM and standard GCN models. Forecasts closely aligned with actual labour demand values ranging from 98 to 148 across a 12-month evaluation period. The proposed model offers policymakers, educational planners, and employers a powerful data-driven tool for workforce planning, skills development, and strategic policy design in an era of continuous digital transformation.