Generalized Self-Attentive Spatiotemporal GCN with OPTICS Clustering for Recommendation Systems

In today's data-driven world, recommender systems are essential for filtering information to deliver personalized content, with collaborative filtering (CF) being a popular technique for predicting user preferences based on past interactions. However, capturing the temporal dynamics of user behavior and handling cold-start scenarios remain significant challenges, as user preferences naturally evolve over time and CF often relies solely on historical data. To address these issues, we propose a novel approach that combines a self-attention-based spatiotemporal graph convolutional network with OPTICS clustering which forms adaptive, time-sensitive subgraphs. This enables the model to adapt to changing user preferences and prioritize the most relevant interactions dynamically. Evaluated on the MovieLens100k dataset, our model outperforms baseline methods, effectively generating embeddings for new user-item pairs and demonstrating an inductive capability that enhances both accuracy and adaptability in recommendations.