Adaptive Ensemble Probabilistic Matrix Approximation for Recommendation
Matrix approximation has been increasingly popular for recommender systems,which have achieved excellent accuracy among collaborative filtering methods.However,they do not work well especially when there are a large set of items with various types and a huge number of users with diverse interests.In this case,the complicated structure of sparse rating matrix introduces challenges to the single global or local matrix approximation.In this paper,we propose an Adaptive Ensemble Probabilistic Matrix Approximation method(AEPMA),which can potentially alleviate the data sparsity and improve the recommendation accuracy.By integrating the global information over the entire rating matrix and local information on subsets of user/item ratings in a stochastic gradient boosting framework,AEPMA has the ability to capture the overall structures information and local strong associations in an adaptive weight strategy.A series of experiments on three real-world datasets(Ciao,Epinions and Douban)have shown that AEPMA can effectively improve the recommendation accuracy and scalability.
Adaptive Ensemble Global and Local Matrix Approximation Matrix approximation
Xingxing Li Liping Jing Huafeng Liu
Beijing JiaoTong University,Beijing 100044,China
国际会议
广州
英文
328-339
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）