Research On Predicting Positive And Negative Links In Social Networks By Ensemble Learning

In recent years, with the development of online social networks and advancement of computer processing ability, analysis and mining of social networks has become a hot research field. Most of research on social networks considers only positive links, we study signed social networks with both positive and negative links. In such networks, a positive link indicates a positive attitude such as friendship and trust, whereas a negative link expresses a negative attitude such as hostile and distrust. An important issue in these networks is to predict positive and negative links, which has many applications, such as friendship recommendation and speculation of users' attitude.In order to improve performance of predicting positive and negative links in signed social networks, we propose an ensemble learning based method. Compared to existing supervised learning based approaches, our method improves from three aspects. First, in feature vector construction phase, we extend existing features to capture more useful information. Second, the distribution of positive and negative links in signed social networks is imbalanced, with much more positive links than negative ones. We address the class imbalance problem and introduce k- means clustering based under-sampling methods to balance the classes. Finally, unlike existing supervised learning based methods that use a single classifier to predict, we take the advantage of ensemble learning and integrate it with under-sampling methods to combine the results of multiple classifiers.To demonstrate the effectiveness of our method, we conduct experiments on two large real-world data sets. The experimental results show that our method improves performance in terms of precise, recall, F1- measure and AUC(Area Under roc Curve).