Combination of Dimensionality Reduction and User Clustering for Collaborative-Filtering
Ngo Tung Son, Dao Huy Dat, Nguyễn Quang Trung, Bùi Ngọc Anh · 2017
In recent years, the Recommender System has been a promising prospect. Many studies show that Collaborative Filtering is commonly used in successful systems. However, these systems are more and more face the problem of large data and sparsity. It brings issues of computation cost as well as low quality of the recommendations. Traditional collaborative filtering methods use the K-nearest neighbor model for generating recommendations. It requires the computation on the whole space. Hence the cost is very high. Moreover, large systems requires the execution process to be processed in parallel that the traditional collaborative filtering does not meet. This paper reports an empirical work related to the application of the combination between Dimension Reduction and User Clustering in Collaborative Filtering. We use available algorithms as the pre-processing phase before applying traditional methods to overcome these problems. Our results bring value to Collaborative-Filtering based system that processing data on big-data ecosystems. The experiments show results in both the processing time of the system and the accuracy of the method.