Personalized product recommendation system for e-commerce platforms

Abstract

Due to rapid growth in e-commerce, the interest for customized product recommendation systems has grown a lot with high demands for effective models. The attempt is made to explore the development and evaluation of a personalized product recommendation model using the H&M data set. The research highlights the building up of an interaction matrix between a user and items, generation of recommendations suited to the tastes of a particular user, and the hyperparameter tuning of the model for better performance. Different techniques have been utilized, including KNNBasic, Non-negative Matrix Factorization (NMF), CoClustering, and Singular Value Decomposition (SVD). The KNNBasic model had a root mean square error (RMSE) of 0.5022 with an accuracy of 42.00%, the NMF model showed better results with an RMSE of 0.4999 and accuracy of 51.50%. Co-Clustering showed the result of RMSE as 0.5000 and accuracy was 50.50%. Notably, the final SVD model ranked very well compared to the others with an RMSE 0.2261 and a great accuracy of 90.40% in this experiment, emphasizing the importance of advanced techniques in recommendation systems. In these experiments, not only is the relative efficacy of different recommendation algorithms evident but also that optimization of hyperparameters genuinely contributes to increasing predictive precision