Research On RBM-based Collaborative Recommendation Algorithms

The rapid development of the era of big data and the era of intelligence has prompted the Recommender System(RS)to gradually become the "standard configuration" of commercial applications.Since the introduction of the Restricted Boltzmann Machine(RBM),due to its structural flexibility and the maturity of related learning algorithms,it has played a role in many tasks in many fields.But research is limited to a few issues in collaborative recommendations and is not sufficient on other issues.Based on the applications of the RBM in the recommendation field,this thesis studies several important issues of RS:(1)Hybrid One-Class Collabortive Filtering(OCCF)based on RBMs.For the OCCF problem,we explore the complementarity between the RBM models and other classical models.Specifically,we try and use two hybrid methods to further improve the recommendation effect;(2)Research on Collaborative Ranking(CR)recommendation algorithms based on RBMs.For the CR problem,we explore the effective use of the RBM models on this problem.Taking the fine granularity of the training data of the CR problem as a breakthrough,several ideas for viewing the data are proposed,and the corresponding recommendation algorithms based on the Conditional RBM(CRBM)is designed.(3)Based on the CRBMs,research on Heterogeneous OCCF(HOCCF)algorithms.For the HOCCF problem,a model based on CRBM is proposed,which not only pays attention to the recommendation effect,but also focuses on the application efficiency.In the research on the above problems,we carried out a large number of experiments to verify the effectiveness of the used or proposed algorithms.Based on these empirical studies,we have obtained some useful conclusions.