A Cross-Domain Recommendation Algorithm for D2D Multimedia Application Systems

IEEE Access

Xu Yu¹  Lingwei Xu¹  Guozhu Liu¹  Yu Fu¹ 

[1] School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, China;

关键词: Cross-domain recommender system;    Funk-SVD model;    imbalanced classification;    transfer learning;   

DOI  :  10.1109/ACCESS.2018.2876885

来源: DOAJ

【 摘 要 】

Device-to-device communication is the key technology of the fifth generation mobile communication, which can support mobile multimedia applications. In recent years, mobile multimedia applications, however, have produced a lot of data, which leads to severe information overload. So recommendation models for multimedia application system have become very popular, which aim to overcome the information overload problems by inferring user preference according to user behaviors, such as rating matrices. However, rating matrices are very sparse and skewed since users are always unwilling to rate items, especially those they do not like. Previous recommendation algorithms cannot address the sparsity and skewed distribution challenges effectively. We propose a cross-domain recommendation algorithm based on feature transfer and imbalanced classification in this paper. First, the original recommendation problem is formulated as a rough imbalanced classification problem in the target domain, which takes user and item location as the feature vector and their rating as the label. Then, useful user and item features are generated to alleviate the sparsity in the target domain. More specifically, extra user features are extracted with a Funk-SVD model from use-side auxiliary domains, and item features are retrieved from Wikipedia. Finally, an imbalanced classification model (AdaBoost.NC) is employed to solve the obtained imbalanced classification problem, which can effectively overcome the skewed rating distribution. We conduct extensive experiments and compare the proposed algorithm with various state-of-the-art single- and cross-domain recommendation algorithms. The experimental results show that the proposed algorithm has advantages in terms of four different evaluation metrics.

【 授权许可】

Unknown