Abstract
Learning user/item relation is a key issue in recommender system, and existing methods mostly measure the user/item relation from one particular aspect, e.g., historical ratings, etc. However, the relations between users/items could be influenced by multifaceted factors, so any single type of measure could get only a partial view of them. Thus it is more advisable to integrate measures from different aspects to estimate the underlying user/item relation. Furthermore, the estimation of underlying user/item relation should be optimal for current task. To this end, we propose a novel model to couple multiple relations measured on different aspects, and determine the optimal user/item relations via learning the optimal way of integrating these relation measures. Specifically, matrix factorization model is extended in this paper by considering the relations between latent factors of different users/items. Experiments are conducted and our method shows good performance and outperforms other baseline methods. Copyright © 2015 Springer International Publishing Switzerland.
Original language | English |
---|---|
Title of host publication | Advances in knowledge discovery and data mining: 19th Pacific-Asia Conference, PAKDD 2015, Ho Chi Minh City, Vietnam, May 19-22, 2015, proceedings, part I |
Editors | Tru CAO, Ee-Peng LIM, Zhi-Hua ZHOU, Tu-Bao HO, David CHEUNG, Hiroshi MOTODA |
Publisher | Springer |
Pages | 732-743 |
ISBN (Print) | 9783319180311 |
DOIs | |
Publication status | Published - 2015 |