Item orientated recommendation by multi-view intact space learning with overlapping

• Published in: Knowledge-based systems Vol. 164; pp. 358 - 370
• Main Authors: Hu, Qi-Ying, Huang, Ling, Wang, Chang-Dong, Chao, Hong-Yang
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 15.01.2019; Elsevier Science Ltd
• Subjects: Algorithms; Intact space; Item orientated; Multi-view; Overlapping; Recommendation; Recommender systems; Representations; Visualization; Overlapping; Multi-view; Intact space; Recommendation; Item orientated
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2018.11.005

Item orientated recommendation has recently been proposed to address the issue of finding the most suitable users of certain item so as to maximize the revenue of its manufacturer with limited advertising budget, which cannot be settled very well by the traditional recommendation algorithms. Existing multi-view models are instructive to solve this problem but either need extra information or suffer from scalability issue due to the high complexity when dealing with large number of records. This paper presents a solution to meet the need of improving both the quality and the scalability of item orientated recommendation by designing a new multi-view model. The basic idea is to visualize the rating records of each user as feature representations from an individual view to partially characterize the property of items. From the multiple but insufficient view representations, the latent feature representation of each item in the intact space, the view generation function of each user, and the overlapping of preferences among different users can be discovered, which are used to predict the unknown ratings from users to a certain item. Extensive experiments have been conducted on three types of datasets, and the results have confirmed the effectiveness of the proposed algorithm. •This paper proposes a new item orientated recommendation algorithm.•It captures the latent feature vector of each item in the intact space.•It captures the differences and overlapping preferences among different users.•Extensive experiments have been conducted to show the effectiveness.