Poster: on trust evaluation with missing information in reputation systems

Reputation plays a critical role in managing trust in decentralized systems. Quite a few reputation-based trust functions have been proposed in the literature for many different application domains. However, one cannot always obtain all information required by the trust evaluation process. For example, access control restrictions or high collect costs might limit the ability gather all required records. Thus, one key question is how to analytically quantify the quality of scores computed using incomplete information. In this paper, we start a first effort to answer the above question by studying the following problem: given the existence of certain missing information, what are the worst and best trust scores (i.e., the bounds of trust) a target entity can be assigned? We formulate this problem based on a general model of reputation systems, and examine the monotonicity property of representative trust functions in the literature. We show that most existing trust functions are monotonic in terms of direct missing information about the target of a trust evaluation.