标题：Review-Based Cross-Domain Recommendation Through Joint Tensor Factorization
作者：Song, Tianhang; Peng, Zhaohui; Wang, Senzhang; Fu, Wenjing; Hong, Xiaoguang; Yu, Philip S.
作者机构：[Song, Tianhang; Peng, Zhaohui; Fu, Wenjing; Hong, Xiaoguang] Shandong Univ, Sch Comp Sci & Technol, Jinan, Shandong, Peoples R China.; [Wang, Senzh 更多
会议名称：22nd International Conference on Database Systems for Advanced Applications (DASFAA)
会议日期：MAR 27-30, 2017
来源：DATABASE SYSTEMS FOR ADVANCED APPLICATIONS (DASFAA 2017), PT I
关键词：Cross-domain recommendation; Joint tensor factorization; Review text
摘要：Cross domain recommendation which aims to transfer knowledge from auxiliary domains to target domains has become an important way to solve the problems of data sparsity and cold start in recommendation systems. However, most existing works only consider ratings and tags, but ignore the text information like reviews. In reality, review text in some way well explains the reason why a product could gain such high or low ratings and reflect users' sentiment towards different aspects of an item. For instance, reviews can be taken advantage to obtain users' attitudes towards the specific aspect "screen" or "battery" of a "cell phone". Taking these aspect factors into cross domain recommendation will bring us more about user preference, and thus could potentially improve the performance of recommendation. In this paper, we for the first time study how to fully exploit the aspect factors extracted from the review text to improve the performance of cross domain recommendation. Specifically, we first model each user's sentiment orientation and concern degree towards different aspects of items extracted from reviews as tensors. To effectively transfer the aspect-level preferences of users towards items, we propose a joint tensor factorization model on auxiliary domain and target domain together. Experimental results on real data sets show the superior performance of the proposed method especially in the cold-start users in target domain by comparison with several state-of-the-arts cross domain recommendation methods.