Design and Implementation of a Hybrid Recommender Model in E-Commerce Systems

Abstract

In this study, a recommender system based on matrix factorization was designed and implemented for an e-commerce platform to provide accurate and personalized predictions based on users’ interactions with products. The primary objective of the research was to improve recommendation accuracy and reduce prediction error by integrating collaborative filtering and content-based algorithms. The dataset comprised user interactions with 67 products and 31 users across 301 shopping carts, which were preprocessed and divided into training and testing sets. The matrix factorization algorithm was applied with optimization of parameters including the number of iterations and the approximation rank, and its performance was evaluated using accuracy, recall, F1 score, prediction error, coverage, and diversity metrics. The results indicated that the model achieved an accuracy of 92% and a prediction error of 0.48, outperforming conventional collaborative filtering and content-based approaches in predicting user preferences and delivering diverse recommendations. Parameter optimization, algorithm integration, and interactive data analysis enabled the model to demonstrate stable and reliable performance. The study’s limitations include the relatively small dataset and the model’s dependence on historical data, which may be addressed in future research through larger datasets, hybrid methods, reinforcement learning, and graph-based models. Practical applications of the model include recommender systems for online retail stores, educational platforms, music and video services, social networks, and targeted advertising, contributing to increased user satisfaction and loyalty, optimized shopping experiences, and the delivery of personalized recommendations.