Like-tasted user groups to predict ratings in recommender systems

Recommendation systems have gained the intention of many researchers due to the growth of the business of personalizing, sorting and suggesting products to customers. Most of the rating prediction in recommendation systems are based on customer preferences or on the historical behavior of similar customers. The similarity between customers is generally measured by the number of times customers liked or disliked the same item. Given the huge number and the variety of items, many customers cannot be considered as similar, as they did not evaluate the same items, even if they have similar tastes. This paper presents a new method of rating prediction in recommendation systems. The proposed method starts by identifying the taste directions or the interest centers based on the users’ demographic information combined with their previous evaluations. Thus, it uses the principal component analysis to retrieve the major taste orientations. According to these orientations, user groups are created. Then, for each group, it generates a prediction model that will be used to predict unknown rates of users within the corresponding group. In order to assess the accuracy of the proposed method, we compare its results with four baseline methods, namely: RegSVD, BiasedMF, SVD++ and MudRecS. The results prove that the proposed algorithm is more accurate than the baseline algorithms.