Research On Trust-Oriented Rating Model On Pure Rating System

In Internet, it's a usual way for people to know the unknown objects by using others'reference, but he needs to know how much he can trust the authors of the ratings. To judge the trust relationship, traditional network security technology can be used, such as digital certificate, however it's short of scalability.With an open rating system, more ratings will appear when everyone can give his own rating. This model has good scalability but the inquirer still needs to know how much he can trust the author.Some algorithms have been worked out to solve the problem in trust-network, and most of the studies are in the field of e-business system or P2P file sharing system. In these kinds of systems, trust relationship between nodes/users can be given directly or be computed by the interactivity between nodes/users.Another kind of system is also very popular, such as rating system about books, which has some diffenrence with the upper system: only ratings from user to object exist and no ratings between users is in it. We call it pure rating system. In this kind of system, it's difficult to get the trust relationship between users. The target of this papar is to work out a model based on trust computing in pure rating system and when some user inquires a target object, this model can be used to find reliable users and synthesize their ratings on the target and finally get reasonable synthesized rating.Aiming at the attribute of the pure rating system– there is no any interactivity between users and obtaining direct trust degree is difficult, this paper brings forword this viewpoint: the"Trust"in pure rating system should be how much one user's rating is accepted by another user. So the nearness degree of ratings between users is used to compute direct trust degree.Based on this viewpoint, this paper proposes a basic model on the assumption that the majority of people in the system are reliable and the ratings between them are near. When some user inquires about an object, this basic model will find the users whose ratings are supported by most people and use their ratings as reference. The result of simulation experiments show that this algorithm can reduce computing errors and solve slandering.But the former assumption does not always come into existence, so this paper proposes an improved model: first find the users whose ratings are near the inquirer (directly or indirectly) and use their ratings as reference to get the synthesized rating which is near the inquirer's viewpoint. So this improved model can also work well when the reliable users are not in the majority, etc. The improved model also considers how to do optimizing when the direct trust degree has comparatively big error. Simulation experiments also.show that this model is available and more efficient than basic models in some conditions.Simulation prototype on pure rating system is also workded out to show how it works in practice.