Sampling and clustering multipile data sources for recommender system

Abstract

Recommending relevant items of interest for user is function of recommender system based on long-term user profiles. However, personal data privacy is becoming a big challenge recently. Thus, recommender system needs to reduce the dependence on user profiles while still keep high accuracy on recommendation. Session-based recommendation is a recently proposed approach for recommender system to overcome the issue of user profiles dependency. In many real-world applications, session-based recommendation is made merely based on last clicked item. Since the relevance of problem is quite high, it has triggered interest among researchers to propose algorithms aiming to create short-term prediction of user’s behavior. In this thesis, we would like to compare the results of such algorithms by using various datasets and evaluation metrics. The most recent deep learning approach named GRU4REC [1] and other wide used approaches are included in our comparison. This thesis also proposes a clustering mechanism and a sampling mechanism fromthe deep learningmodel for better performance. Six real-world datasets from three different domains were experimented to evaluate whether high performance of GRU4REC can achieve cross-domains. This work also revealed that this deep learning algorithm achieving low result on some specific datasets.