GASP: A Graph Augmentation-Based Approach for Sign Prediction of Ties in Social Networks

This paper proposes a method for the sign prediction of ties in social networks using a design science research process. The proposed method is grounded in social network analysis and leverages the tenets of graph augmentation and a graph regularized framework for information diffusion. To the best of our knowledge, this is the first study that develops a sign prediction method for social networks based on the principles of design science research. This study makes several contributions. We demonstrate the utility and applicability of the proposed method for predicting trust/distrust on a user-user network created from the IMDb platform, which represents ties between reviewers based on their movie evaluation preferences. We describe and discuss the novel aspects of graph augmentation, symmetric normalization of the affinity matrix, and graph regularized label propagation, and discuss their synergistic use to predict the signs of network ties. We also establish the effectiveness of the proposed method by comparing its performance with two different metrics for balanced networks and two metrics for unbalanced networks using four state-of-the-art methods. The benchmarking networks used for experiments originate from online platforms such as Slashdot, Epinions, Wikipedia, and the Yeast Genetic Interaction Network from the biology domain. Experiments show that the proposed method provides significant performance improvements in the sign prediction of ties in social networks. This study provides valuable insights for social media platform owners seeking to improve their platforms by building new features and business leaders seeking to target advertisements and personalized content to users. We also discuss the theoretical, practical, and societal implications of this research.