Knowledge-Based Systems 23(2010)232-238 Contents lists available at Science Direct Knowledge-Based Systems ELSEVIER journalhomepagewww.elsevier.com/locate/knosys Improved trust-aware recommender system using small-worldness f trust networks Weiwei Yuan, Donghai Guan Young-Koo Lee a,, Sungyoung Lee Sung Jin Hur b Republic of Kore b Electronics and Telecommunications Research Institute(ETRI), Republic of Korea ARTICLE INFO A BSTRACT The trust network is a social network where nodes are inter-linked by their trust relations. It has bee idely used in various applications, however, little is known about its structure due to its highly dy Received in revised form 23 November 2009 pted 31 December 2009 nature. Based on five trust networks obtained from the real online sites, we contribute to verify that the Available online 6 January 2010 rust network is the small-world network: the nodes are highly clustered, while the distance between two randomly selected nodes is short. This has considerable implications on using the trust network in the trust-aware applications. We choose the trust-aware recommender system as an example of such applications and demonstrate its advantages by making use of our verified small-world nature of the 2010 Elsevier B.V. All rights reserved. Recommender system 1 Introduction its trust on any user of the trust network. This irregular growth leads to the complex structure of the trust network. Since the topology of The trust-aware recommender system (TARS)is the trust network is the important information to optimize TArs. mender system that suggests the worthwhile informat this research motives to make clear the structure of the trust net sers on the basis of trust, in which trust is the measu work. Furthermore based on the topology of the trust network, we less to believe in a user based on its competence an motive to optimize the conventional TARS model within a specific context at a given time. tars has re The contributions of this paper are mainly in two-fold proposed for use since it is able to solve the well-known data arseness problem of the collaborative filtering(CF)(1, 2]. This is We conduct experiments to verify the small-world topol- because trust is transitive. It means. if a trusts b and b trusts C.a ogy of the trust network, which can facilitate its usage in trusts c to some extend so even if there is no direct trust between arious trust-aware applications. Though the trust network the active users and the recommenders the active users can build has been assumed to be a small-world network by some up some indirect trust relationships with the recommenders via existing works 3-5, to the best of our knowledge, no the trust propagations. This contributes to the high rating predic one has verified its small-worldness experimentally or the- tion coverage of TARS. Moreover, the rating prediction accuracy etically. By analyzing five trust networks extracted from of TARS is no worse than the classical CF 1 e real online sites, we contribute to verify that the trust Despite of its high rating prediction accuracy and high rating pre- network is the small-world network: on one hand. the diction coverage, the conventional TARS model suffers from the nodes of the trust network are highly clustered, which is problem that it is not optimized: its computational complexity can similar to the regular network; on the other hand, the dis- be exponentially more expensive by achieving similar rating predi- tance between two randomly selected nodes of the trust cation accuracy and rating prediction coverage, and its rating predic network is short which is similar to the random network. tion coverage can be significantly worse by achieving similar rating We propose a novel TARS model which can effectively predication accuracy. This is because little is known about the topol- vercome the weakness of the conventional tars model ogy of the trust networks used in TARS. The trust network is highly This is achieved by leveraging our verified small-worldness ynamic: a user can join the trust network at anytime by stating ntal results clear show that. our proposed model is superior to the conventional one since it is able to achieve the maximum rating prediction accuracy and the maximum rating prediction coverage vith the minimum computat 0950-7051/s-see front matter o 2010 Elsevier B V. All rights reserved o:10.1016/ knosys2009.12004Improved trust-aware recommender system using small-worldness of trust networks Weiwei Yuan a , Donghai Guan a , Young-Koo Lee a,*, Sungyoung Lee a , Sung Jin Hur b aDepartment of Computer Engineering, Kyung Hee University, Yongin, Republic of Korea b Electronics and Telecommunications Research Institute (ETRI), Republic of Korea article info Article history: Received 2 July 2009 Received in revised form 23 November 2009 Accepted 31 December 2009 Available online 6 January 2010 Keywords: Trust Trust network Small-world network Recommender system abstract The trust network is a social network where nodes are inter-linked by their trust relations. It has been widely used in various applications, however, little is known about its structure due to its highly dynamic nature. Based on five trust networks obtained from the real online sites, we contribute to verify that the trust network is the small-world network: the nodes are highly clustered, while the distance between two randomly selected nodes is short. This has considerable implications on using the trust network in the trust-aware applications. We choose the trust-aware recommender system as an example of such applications and demonstrate its advantages by making use of our verified small-world nature of the trust network. 2010 Elsevier B.V. All rights reserved. 1. Introduction The trust-aware recommender system (TARS) is the recom￾mender system that suggests the worthwhile information to the users on the basis of trust, in which trust is the measure of willing￾ness to believe in a user based on its competence and behavior within a specific context at a given time. TARS has recently been proposed for use since it is able to solve the well-known data sparseness problem of the collaborative filtering (CF) [1,2]. This is because trust is transitive. It means, if A trusts B and B trusts C, A trusts C to some extend. So even if there is no direct trust between the active users and the recommenders, the active users can build up some indirect trust relationships with the recommenders via the trust propagations. This contributes to the high rating predic￾tion coverage of TARS. Moreover, the rating prediction accuracy of TARS is no worse than the classical CF [1]. Despite of its high rating prediction accuracy and high rating pre￾diction coverage, the conventional TARS model suffers from the problem that it is not optimized: its computational complexity can be exponentially more expensive by achieving similar rating predi￾cation accuracy and rating prediction coverage, and its rating predic￾tion coverage can be significantly worse by achieving similar rating predication accuracy. This is because little is known about the topol￾ogy of the trust networks used in TARS. The trust network is highly dynamic: a user can join the trust network at anytime by stating its trust on any user of the trust network. This irregular growth leads to the complex structure of the trust network. Since the topology of the trust network is the important information to optimize TARS, this research motives to make clear the structure of the trust net￾work. Furthermore, based on the topology of the trust network, we motive to optimize the conventional TARS model. The contributions of this paper are mainly in two-fold: – We conduct experiments to verify the small-world topol￾ogy of the trust network, which can facilitate its usage in various trust-aware applications. Though the trust network has been assumed to be a small-world network by some existing works [3–5], to the best of our knowledge, no one has verified its small-worldness experimentally or the￾oretically. By analyzing five trust networks extracted from the real online sites, we contribute to verify that the trust network is the small-world network: on one hand, the nodes of the trust network are highly clustered, which is similar to the regular network; on the other hand, the dis￾tance between two randomly selected nodes of the trust network is short, which is similar to the random network. – We propose a novel TARS model which can effectively overcome the weakness of the conventional TARS model. This is achieved by leveraging our verified small-worldness of trust networks. Experimental results clear show that: our proposed model is superior to the conventional one since it is able to achieve the maximum rating prediction accuracy and the maximum rating prediction coverage with the minimum computational complexity. 0950-7051/$ - see front matter 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.knosys.2009.12.004 * Corresponding author. E-mail address: yklee@khu.ac.kr (Y.-K. Lee). Knowledge-Based Systems 23 (2010) 232–238 Contents lists available at ScienceDirect Knowledge-Based Systems journal homepage: www.elsevier.com/locate/knosys