PRE-ADDL: Optimized Attention-Driven Deep learning Mechanisms for Accurate and Computationally Efficient E-Commerce Recommendations

Abstract

: In directing users' decision-making across a variety of online platforms, recommendation systems are essential. Enhancing the accuracy and relevance of these systems has become a more significant challenge in both academic research and industry applications as the volume of web data keeps growing. Although many models have been developed to address this issue, the effectiveness of many traditional approaches can be hampered by their reliance on narrow perspectives. Using web usage mining techniques, we present an Attention-Driven Deep Learning Model based recommendation system in this study. In order to provide more precise and tailored recommendations, our method looks for intricate patterns and connections in user behaviour and online interactions. We evaluated our approach on public web log datasets, using a temporal evaluation protocol that simulates the dynamics of an E-commerce website in a realistic way. The study found that although more than 1.4 million users engaged with products, just 0.83% of them became buyers, which indicates the difficulty of enhancing engagement and conversion rates. A deep Learning model utilizing an attention mechanism was built to improve personal recommendations. The architecture of the model involves various layers, i.e., embedding, attention, feature extraction, and dense layers, to effectively capture user-item interactions. Experimental results showed that the model reported approx. 97% accuracy with excellent Precision and Recall. The recommendation system efficiently yielded top 5 product recommendations to users, where relevant items recorded probability scores ranging up to 0.0257. Computational efficiency is revealed through the 0.63 seconds response time. The study's findings highlight how well deep learning can enhance user engagement and streamline personalized suggestions, creating opportunities for additional advancements in e-commerce recommendation systems.