Manipulating recommendation lists by global considerations

The designers of trust and reputation systems attempt to provide a rich setting for interacting users. While most research is focused on the validity of recommendations in such settings, we study means of introducing system requirements and secondary goals which we term Global Considerations. Recommendation systems are assumed to be based on a framework which includes two types of entities: service providers and users seeking services (e.g. eBay (eba, )). The present paper formulates a basis for manipulation of information in a manner which does not harm either. These manipulations must be carefully devised: an administrator attempting to manipulate ratings, even for the benefit of most participants, may dampen the gain of service providers, users or both. On the other hand, such changes may produce a more efficient and user friendly system, allow for the improved initialization of new service providers or upgrade existing features. The present paper formulates threshold values which define the limits of our manipulation, propose different concepts for manipulation and evaluates by simulation the performance of systems which employ our manipulations.