Building robust Reputation Systems for travel-related services

There is a serious robust issue of building Reputation Systems for travel-related services, such as hotel, restaurant, etc. This paper proposes an advanced clustering approach, Suspicion Degree Meter (SDM), to rank suspects with respect to manipulative behavior. The detection process proceeds on different layers, review layer, traveller layer and hotel layer. Regarding two types of manipulative behavior, i.e. promoting and demoting manipulation, SDM assigns two real numbers, Suspicion Index for Promoting and Suspicion Index for Demoting, to each object in different layers. Inherent logical implication among different layers is considered to adjust the original value of Suspicion Index. Sets of suspects in different layers are identified. Furthermore, some practical reputation models are proposed to enhance the robustness of Reputation Systems. In the evaluation phase, statistical character of suspects and innocent subpopulation are compared. Some interesting conclusions and phenomena are shown. Meanwhile, by using a proposed reputation-model-comparison approach, Ranking Variation Analysis, all the reputation models are compared in terms of efficiency against manipulation. One of the most significant conclusions is that, although there is not a universal reputation model which fits best for every circumstance, given a set of suspects identified by SDM, local optimization can be achieved.