A Self-Attention-Driven Deep Learning Framework for House Price Prediction Using Multimodal Data

House price prediction remains a critical challenge in the real estate domain, requiring the consideration of diverse factors that influence housing prices. Traditional models often fail to capture the intricate relationships between these factors, resulting in limited predictive accuracy. To address this, we propose an end-to-end Joint Self-Attention Mechanism for house price prediction, integrating public facility data (e.g., parks, schools, transit) and satellite imagery to assess environmental contexts. Our approach emphasizes the identification of critical features and their interactions, enabling the discovery of complex spatial and contextual relationships.By combining the Joint Self-Attention Mechanism with Spatial Transformer Networks (STN), we aim to establish a robust and precise framework for understanding housing market dynamics. This innovative methodology introduces a new perspective on integrating spatial and contextual data for more informed and reliable house price predictions.