Predicting Implicit Negative Relations in Online Social Networks

Abstract

Social network analysis can reveal a lot of information about the users of a network. Both positive and negative links in a social network can be useful to analyze and predict relationships between the users of a social network. Although there has been a lot of research on positive link prediction, there is still a scarcity of research on negative link prediction. We exploit this gap by predicting both positive and negative links within an online social network. Negative link prediction can help us to detect an implicit negative relationship in a social network. Feelings such as hatred, animosity, and antipathy can be useful to reveal facts such as whom to avoid or not suggest as a possible friend. Also, predicting unknown links between users can be used to design a promising recommendation system. This paper makes use of the Epinions and Slashdot dataset to train the model and detect implicit negative links in any social network by using a logistic regression-based technique. Using the selected feature sets, the results obtained have a good precision of detecting an unknown link.