1 UIP: Estimating True Rating Scores of Services through Online User Communities Xiaodi Huang Charles Sturt University, Australia xhuang@csu.edu.au Weidong Huang University of Tasmania, Australia tony.huang@utas.edu.au Wei Lai Swinburne University of Technology wlia@swin.edu.au Abstract— As many online systems rely on user ratings for making decisions such as recommendations, the quality of such rating scores are increasingly important. On the other hand, users interact with each other via online communities. How such interactions affect the trueness of their ratings? Can we obtain the true rating scores that exclude the influences among users? This paper presents a conceptual framework that characterizes the influences on quality of services among users, and an algorithm that estimates the true rating scores by minimizing the influence among users. In other words, the influence on users’ ratings due to their interactions is minimized so as to obtain the more accurate rating scores. The proposed approach has been validated by experimenting on real data sets. The results of the experiments have demonstrated that our approach is capable of estimating true ratings. Keywords—services; online social network; ratings; propagation I. INTRODUCTION Relying on ratings by users is a very common approach to assessing quality of online services [14,15]. With the proliferation of online services, the estimation of such accurate ratings has become a critical issue for service providers[16]. In reality, users can easily share their opinions or experience of using a service and purchasing a product with other community members who have no direct experience with them. As we know, mutual reinforcement takes place among hyperlinks [6, 8]. The Web is evolving into Web 2.0, which is saliently characterized by facilitating collaboration and sharing information between users in an unprecedented way. Web 2.0 provides both the hyperlinks of Web pages and a platform in which users are able to interact with each other. Not only is useful the hyperlinks, but user interactions are also informative. The members of an online user community can interact with each other in various ways such as reading, commenting, posting, and discussing ratings. As a result, information and influence are spread among its members. Approaches that incorporate social influence into recommender systems or online marketing in E-commerce are capable of improving the performance. They, however, focus on the influence on recommending products. It seems to be increasingly important to improve the accuracy of estimating the ratings scores of services by measuring online social influence. In a word, link analysis ranking makes use of the hyperlink relations among Web pages, while Web 2.0 makes it possible to exploit the interactive information among a set of users in order to indirectly quantify the quality of services. In this paper, our focus is to explore the extent to which user ratings are affected by their peers from the same online community. The basic idea of our approaches is to employ indirect computations. Relying on a user social network, the indirect approach makes use of the influence mechanism so as to indirectly measure the quality of services. In order to obtain the true rating score of a service, we take into consideration the influence between users with respect to their evaluations on services. To the best of our knowledge, this is the first work on integrating social influence on user ratings into measurement in social networks for estimating the true quality of services. In particular, the contributions of this research are twofold. —We describe a new conceptual model that characterizes the influence among users on rating the quality of a service. —We present an approach to compute user influences on ratings and quality scores of services, together with the experiment results. The remainder of this paper is organized as follows. Related work is given in the following section. A conceptual model for services is presented in Section 3. Section 4 presents the preliminaries and the algorithms for computing the true ratings of services, followed by reporting experiments in Section 5. Section 6 concludes this paper.. II. RELATED WORK In this section, we review the relevant literatures on influences in online communities and estimations of QoS parameters, as well as compare our algorithm with other relevant ones. The recent proliferation of online networks has aroused much research interests in computer science. Roughly, there are two categories of relevant research on online networks. One area is in characterizing an online community itself and its members, such as discovery of communities [2, 4, 9], community evolution [7], and pattern mining in network data. The interests of users within a social network are inferred using social neighbors [12]. Modeling the dynamics and topic dependency of social influence, Tang et al. [11] measures the topic-level social influence on large-scale networks. Building upon social influence and correlation in online communities, the improvement on the quality of services or recommendations is the second area. Social network researchers