A comparison of reputation-based trust systems

John B. Folkerts

Physical copy available from RIT's Wallace Library at BF575.T7 F65 2005


Recent literature contains many examples of reputation systems which are constructed in an ad hoc way or rely upon some heuristic which has been found to be useful. However, comparisons between these reputation systems has been impossible because there are no established methods of comparing performance. This paper introduces a simulation framework which can be used to perform comparison analysis between reputation models. Two reputation models, one from Abdul-Rahman and Hailes (ARH) [1], and one from Mui, Mohtashemi, and Halberstadt (MMH) [17] are implemented and compared with regard to accuracy, performance and resistance to deception. In order to improve performance in certain cases, MMH is modified to distinguish the concept of "trust" from the concept of "reputation." Additionally we examine the results of shortening the memory of MMH in order to improve results in environments that are constantly changing.