Cluster-Aware Next-Basket Recommendation Using a Hybrid Autoencoder-RNN and Graph Transformer Approach

One of the toughest issues facing contemporary companies is still knowing consumer behavior. Big data's emergence has hastened the development of customer relationship management, allowing businesses to acquire previously unattainable knowledge on consumer preferences and behavior. By means of thorough data analysis, companies can improve consumer happiness and loyalty by means of identification of buying trends and future behavior, thereby refining their marketing initiatives. In the retail and e-commerce sectors, next-basket recommendation (NBR) has grown to be a necessary instrument. Experts have proposed several approaches to improve its efficiency. Although the PIF-kNN technique fails to identify consecutive purchase behavior, it effectively generates suggestions by utilizing tailored item frequency. Although Sets2Sets requires substantial data and computational resources, it employs an RNN encoderdecoder model to capture basket-level buying patterns and repetitions within an end-to-end learning framework. On the other hand, SHAN uses a special attention system to tell the difference between what users like over a long time and what they want in the short term, which improves how accurate the recommendations are, even though the system is quite complex. This work shows a combined method that uses a mix of autoencoder and RNN to understand user behavior and track purchase patterns, along with Graph Attention Networks (GAT) and Transformer models to analyze complicated item connections based on KMeans clustering, aiming to fix the weaknesses of current models. Moreover. The proposed NBR framework undergoes evaluation using publicly available data. This cluster-driven approach customizes results to fit individual user preferences and activity recommendation quality. Evaluation results show that in both accuracy and relevance, it exceeds conventional recommendation systems.