Abstract—Even though recommendation systems have achieved great success both in commerce and academy, there is still much to improve in cross-domain recommendation field. In this paper, we propose a novel framework for cross-domain recommendation in folksonomies: CRF. The idea of CRF is generating user’s tag-profile in the target domain, based on the correlation of tags between different domains. Then the cross-domain issue is transferred into traditional single domain recommendation problem. Compared to related work, CRF is more flexible and scalable, it can adapt to multi-cross-domain recommendation issues. As it is a framework, CRF can be implemented in various ways according to practical requirements. Moreover, CRF is based on folksonomy, so it can be widely used in various applications of Web 2.0. In addition, data sets from previous work are far from satisfaction, so we build a cross-domain data set for evaluation. We validate different realizations of CRF and demonstrate its effectiveness. The test results show that when we choose typical tags as features, the algorithm performs the best. The experiments also show that CRF is more precise than one-domain recommendation algorithms to solve cold-start problem in the target domain.

Index Terms—data mining, cross-domain recommendation, folksonomy, collaborative filtering.


Cite: Ying Guo and Xi Chen, "A Framework for Cross-domain Recommendation in Folksonomies," Jounal of Automation and Control Engineering, Vol. 1, No. 4, pp. 326-331, Dec., 2013. doi: 10.12720/joace.1.4.326-331