Comparing MLP, SVM and KNN for predicting trust between users in Facebook

Trust is one of the most important elements in social relationships. This importance becomes more conspicuous when the person´s interactions take place in a dynamic environment with high uncertainty. Online social networks are websites where information propagates quickly and once the user has entered, they face a huge amount of information all of which they cannot assess. Therefore, it should be possible to rank the presented information based on issues such as trust. Trust prediction in social networks which do not support implicit rating mechanisms is a challenging problem. It is shown in this paper how it is possible to measure the trust between users in Facebook based on the information of their interactions and profile. For this, a dataset comprising interactions, profile information, and trust amount between users was collected. Then after the pre-process of this data set, three methods MLP¹, KNN² and SVM³ were used for trust prediction. Using 10-fold cross validation, we observed that MLP can measure trust with 83% accuracy. The accuracy of the best KNN model, with k=12, and SVM were 73% and 71% respectively. Recall, precision and area under the ROC curve of SVM and KNN were also significantly lower than MLP. According to these results, MLP can measure the trust between users with high accuracy based on the information of their interactions and profile.