Attentive Matrix Factorization for Recommender System

Abstract

The traditional collaborative filtering algorithm is the most common algorithm in the recommender system. Among them, the matrix factorization technique is widely used because of its simplicity and effectiveness. Although matrix factorization effectively solves the problem of data matrix sparsity, it is not enough to use the inner product of the user feature vector and the item feature vector as the user's rating of the target item. This simple linear calculation ignores the user's attention to the different attributes of each item. Based on the matrix factorization, this paper adds an user attention to better analyze the user's attention to the item and obtain the user's preference more accurately. Using RMSE and MAE as evaluation indicators and conducting experiments on a real dataset, the experimental results show that this model is better than other comparison algorithms.