Towards an objective ranking in online reputation systems: the effect of the rating projection

Online reputation systems are commonly used by e-commerce providers nowadays. In order to generate an objective ranking of online items’ quality according to users’ ratings, many sophisticated algorithms have been proposed in the literature. In this paper, instead of proposing new algorithms we focus on a more fundamental problem: the rating projection. The basic idea is that even though the rating values given by users are linearly separated, the real preference of users to items between different values gave is nonlinear. We thus design an approach to project the original ratings of users to more representative values. This approach can be regarded as a data pretreatment method. Simulation in both artificial and real networks shows that the performance of the ranking algorithms can be improved when the projected ratings are used. Introduction. – The coming big data era brings us an critical problem: how to extract the valuable information from the big data at hand. This problem is especially crucial in online systems where the available data are overwhelmingly abundant due to the rapid expansion of the Internet [1–4]. To filter out irrelevant online items (e.g. books, movies or others) for users, the recommender system, such like the collaborative filtering methods are widely applied [5, 6]. Besides the relevance, the quality of items is also of great importance to online users. Therefore, many online websites, such as Amazon.com and Netflix.com build the online reputation system [7–10] in which users can give their opinions to an item by assigning certain rating value to it. The purpose of the reputation system is to help users uncover the true quality of items. After obtaining the rating data, some algorithms are needed to generate the ranking of items. The most straightforward way is to simply use the arithmetic average of ratings to rank items’ quality. However, since this method has low ranking accuracy and is sensitive to spamming behavior, many other ranking algorithms have been proposed recently [11]. Other types of ranking algorithms compute users’ reputation and items’ quality self-consistently. More specifi(a)E-mail: [email protected] cally, these algorithms usually update users’ reputation in an iterative way and aggregate the ratings based on the reputation of users [12]. A representative one of these algorithms is called iterative refinement (IR) [14]. In IR, a user’s reputation is inversely proportional to the mean difference between his rating vector and objects’ estimated quality vector (i.e., weighted average rating based on user reputation). The estimated quality of objects and reputation of users are iteratively updated until the values reach a stationary point. This method is further modified by assigning trust to each individual rating [13, 15]. Recently, Zhou et al. [16] takes the robustness of the algorithm into account and propose to calculate a user’s reputation by the Pearson correlation [17] between his ratings and objects’ estimated quality. This method is usually referred to as the Correlation-based Ranking (CR) method and it can be resistent to the malicious spamming behaviors of some users. However, a fundament problem in the reputation system has been neglected for a long time. For most online reputation systems, the rating values are discrete and linearly separated. For example, some well-known websites such as Amazon.com and Netflix.com use the 5-star rating system: users are allowed to rate items with integers from 1 (worst) to 5 (best) [18,19]. However, the real preference of users to items between different rating values