An attack resistant method for detecting dishonest recommendations in pervasive computing environment

Abstract

An attack resistant method for indirect trust computation (based on recommendation) for pervasive computing environment is proposed. The method extends a mechanism to detect outliers in dataset presented in [19] and apply it to filter out malicious recommendations in indirect trust computation. The method is based on a dissimilarity metric based on the statistical distribution of the recommendations grouped into recommendation classes. The proposed model has been evaluated in different attack scenarios (bad mouthing, ballot stuffing and random attack). The model has also been compared with other existing evolutionary recommendation models in this field, and it is shown that the proposed approach can effectively filter out dishonest recommendations provided that the number of dishonest recommendations is less than the number of honest recommendations.