Employing Locally Linear Neuro-Fuzzy Model for Trust Prediction

Trust network can be useful tool for identification of the trustful sources of information in a social web application. Despite the usefulness of a trust network, it is usually sparse as users are generally reluctant to express the trust relationships. Therefore, one approach to make the trust network denser is to identify the potential trust relationships between users and recommend the reliable and trustful participant to users. The main contribution of this research is the examination of the performance of Locally Linear Neuro Fuzzy (LLNF) model in the trust prediction context. Therefore, LLNF was compared with two classification techniques including ANFIS and C4.5 decision tree. To conduct experiments, data of Epinions network were used. The results of experiments demonstrate that LLNF achieves the highest performance in terms of area under ROC (AUC). In addition, LLNF follows closely C4.5 decision tree regarding F-measure. LLNF can be a good candidate for trust relationships prediction.