Abstract Personalized recommendation systems are critical components in large-scale e-commerce ecosystems, where user engagement and conversion depend heavily on the system’s ability to adapt to diverse behavioral patterns and contextual factors. Conventional approaches, including collaborative filtering and rule-based heuristics, often exhibit limitations in capturing complex user-item relationships, suffer from cold-start issues, and lack responsiveness to temporal context. This paper presents a novel AI-driven hybrid recommendation framework that integrates graph-based relational modeling and deep contextual sequence learning to enhance recommendation accuracy, robustness, and scalability. The proposed architecture leverages Graph Neural Networks (GNNs) to learn latent representations from the user-item bipartite interaction graph, capturing higher-order collaborative signals. In parallel, a Transformer-based encoder processes sequential user interactions enriched with contextual metadata such as timestamp, device type, and location, enabling temporal and situational awareness. A fusion mechanism combines the outputs of both modules to compute relevance scores, which are further refined using a real-time feedback loop incorporating click-through and purchase logs.
Arunkumar Medisetty (Mon,) studied this question.