Handling Sparsity in Cross-Domain Recommendation Systems: Review

Abstract

Cross-domain recommendation Systems (CDRS) is a significant research area which has been target of many companies these days. From last few years, there is amount of publications in CDRS domain including recommendation systems which have been rising sharply alongside information retrieval and machine learning. Recommender systems help companies to specifically identify preferences of the user from the collected data from various sources. These data values are then used to identify user preferences also known as recommendations. In cross-domain, we use the data from one domain such as movies to recommend items in other domains, for instance, Books. There are numerous issues which recommender systems suffer together with sparse data, synonymy, data privacy, algorithm scalability, perspective awareness (context) and cold start problems. Data sparsity is a major issue in recommender systems, especially in the presence of novel users or items, or when user drift exists. This paper reviews recent efforts made for CDRS sparsity and user drift which are prevalent in most CDRSs such as user-based, item-based or knowledge transfer. This paper formalizes the CDRS illustrates sparsity related issues which are addressed in prior works and finally proposes for future research trail.