Trust Analysis Improvement Through Deep Learning for Signed Social Networks

Nowadays, signed social networks are considered the foremost platform for sharing ideas and views by Internet users. Interactions in signed social networks are modeled in the structure of links between two users. These links can have a negative sign indicating distrust and hostility or a positive sign indicating trust and friendship. According to the literature, predicting the sign of links in signed social networks is considered a critical challenge in social network analysis. To address this challenge, several approaches were proposed using two social psychological theories specifically structural balance and social status theories based on the existence of mutual neighbors. Due to the sparseness of signed social networks and the lack of mutual neighbors among users, it is crucial to provide a reliable method for predicting the signs of links. In this study, an attribute vector representing the pair of link nodes is proposed to describe the relevant and latent features of the nodes. Subsequently, a deep learning model based on LSTM is developed to predict the link sign. Finally, the proposed model performance is improved in predicting the link sign. The standard Slashdot dataset is used to assess the proposed method. The results of this study revealed that the proposed method improves the accuracy of the link sign prediction process. Findings lead us to believe that our proposed method is a promising solution in the domain of signed social networks and may increase the reliability of sign prediction as well as its accuracy.