A personalized recommendation method using item domain features to estimate the vacancies of user-item rating matrix

Collaborative filtering (CF) is a widely used personalized recommendation method. However, its performance is often limited by sparsity problem. Some researchers capitalize dimensionality reduction to alleviate this problem, but dimensionality reduction might lose potentially useful information. Although some others fill the vacant ratings by specific technologies to alleviate the sparsity, they never exploit potentially useful information between items to fill the vacancies. In this paper, we introduce a method that utilizes item domain features to estimate the vacant ratings of user-item rating matrix rather than specific technologies, and combines CF to make personalized recommendation. This method not only alleviates the sparsity of CF, but also integrates domain characteristics with a personalized recommendation system. We compare the proposed method with existing methods and find our method gets better performance than other methods.