An Efficient Cross-Domain Recommendation System Based on Clustering Approach and User-Score

Abstract

Growing consumers and online commercial products have put a great challenge on Recommender systems.  Abundant browsing and online shopping across the world have provided immense opportunities to E-commercial giants to effectively learn the behaviour and attract users. However, due to the variety of available products and distinct consumers, crucial issues of data sparsity and cold start problems have been introduced. Single-domain recommendation systems are unable to handle such issues and research has been concentrated on developing more sophisticated cross-domain recommendation systems. Concerning other systems, non-overlapping domains form the most difficult part to solve using the cross-domain approach. This article focuses on partially overlapping domains and suggests a simple and efficient approach based on mapping users' interests by clustering books and movies. The cluster information from the auxiliary domain is transferred to the target domain using the user score and mapped properly to recommend books. The user and item scores in both domains are computed using the transfer frequency-inverse document frequency approach. The proposed recommender system offers low computational complexity and requires less time.