Optimizing Accuracy and Computational intensity of Web Usage Mining for E-commerce Recommendations Using LLMs

Abstract

Web usage mining has become critical in enhancing e-commerce platforms by understanding user behavior and providing personalized recommendations. However, traditional algorithms often suffer from high computational intensity and limited accuracy when processing large-scale web log data. This research proposes a novel approach leveraging Large Language Models (LLMs) to analyse and optimize web log data for e-commerce recommendation systems. To take into account the rich contextual understanding of LLMs and their powerful feature extraction capability, the presented approach is set to enhance precision while reducing computation overhead. By utilizing the E-commerce Website Logs dataset, this research preprocesses raw log data into structured features such as session durations, clickstream patterns, and page dwell times. This approach is then tested against five other existing algorithms: Random Forest, Support Vector Machine (SVM), k-Nearest Neighbours (k-NN), Gradient Boosting, and Deep Neural Networks in terms of accuracy, F1-score, precision, recall, sensitivity and root mean square error (RMSE). Apart from prediction quality, the computational efficiency of each algorithm is also analysed in terms of training time, inference time, and resource usage. Results show that the LLM-based model consistently outperforms traditional algorithms across all the evaluation metrics, such as achieving higher accuracy, precision, and recall along with substantially lower RMSE. In addition, the proposed method has faster processing times and lower computational intensity, which makes it more suitable for large-scale, real-time recommendation tasks. This paper contributes to the field by showing the potential of LLMs in web usage mining and recommendation systems, offering a scalable solution that bridges the gap between accuracy and efficiency. These findings emphasize the integration of advanced models of machine learning, such as LLMs, into e-commerce platforms for the ever-growing demands of today's modern platforms and to ensure computational feasibility.