Time-aware parallel collaborative filtering movie recommendation based on Spark

Abstract

Most of the traditional collaborative filtering (CF) recommendation models mainly consider the users' ratings on items, often ignore the time context information of users. However this information is non-trivial to improve the effectiveness of recommender system. A time-aware parallel CF movie recommendation based on Spark is proposed in this paper. The CF algorithm based on matrix factorisation can associate users' interests with items through implicit features and solve the sparse matrix problem. The time-aware CF algorithm considers the dynamic features associated with the items and users, which improves the recommendation accuracy by introducing discrete time parameter into the matrix factorisation model. To solve the problem of the slow processing speed of high volume data, distributed computing based on Spark is used to achieve the parallelisation of the algorithm. The experimental results on real dataset MovieLens show that the proposed method performs significantly better than traditional CF recommendation, which can alleviate the problem of data sparsity and significantly improve the processing speed and recommendation accuracy.