Grouping Like-Minded Users for Ratings’ Prediction

Abstract

Abstract. Regarding the huge amount of products, sites, information, etc., finding the appropriate need of a user is a very important task. Recommendation Systems (RS) guide users in a personalized way to objects of interest within a large space of possible options. This paper presents an algorithm for recommending movies. We break the recommendation task into two steps: (1) Grouping Like-Minded users, and (2) create model for each group to predict user-movie ratings. In the first step we use the Principal Component Analysis to retrieve latent groups of similar users. In the second step, we employ three different regression algorithms to build models and predict ratings. We evaluate our results against the SVD++ algorithm and validate the results by employing the MAE and RMSE measures. The obtained results show that the algorithm presented gives an improvement in the MAE of about 0.42 and 0.5201 in the RMSE.

Like-tasted user groups to predict ratings in recommender systems

“…The essential thought behind latent factor models is to build a prediction model based on the stored ratings. Once built, the model is used to estimate unknown ratings [14]. Many latent factor models techniques are utilized in the context of CF, such as Matrix Factorization [17], Probabilistic Matrix Factorization [21] and other variations [30].…”