Predicting Interaction Quality in Online Social Networks

The quality of interactions among users on social media sites often depends on both the properties intrinsic to the users themselves as well as on the users’ past interactions on the sites. We study the Chatous random chat network, for which interaction quality is measured by the conversation length between users. By accounting for user properties and modeling user interactions as graph properties, we can accurately predict the quality of interactions between pairs of users in this network before they interact. We experiment with two methods of incorporating user and graph properties. First, we build a support vector machine (SVM) based multiclass classification system by combining user and graph properties as features in a principled manner. Second, we decompose the network into a bipartite graph using the most salient user property gender and implement collaborative filteringand singular value decompositionbased prediction systems. We demonstrate that both approaches produce good results, with 62.2% accuracy for the feature-based classification approach and 59.2% accuracy for the bipartite approach.