Abstract: To overcome the shortcomings of the traditional method in product recommendation, a new recommendation method based on a supervised user rating profile (SURP) model is proposed. The products are represented by their image features using the bag-of-word method. The SURP model is proposed using the supervised learning method, which can effectively improve the precision of the predicted ratings. The interest factor of the users is introduced to deal with the change in the interest of the users on the products bought previously. The method is tested on a product dataset. The experimental results show that the new method has a better recommendation capacity than the user rating profile, Gaussian probabilistic latent semantic analysis, and item-based collaborative filtering methods.