Sentiment-aware competitive graph diffusion model for negative information control in online social networks

To accurately represent diffusion behavior, it is important to know how emotional tone and user credibility affect the spread of information in online social networks. This study introduces a competitive diffusion framework that considers sentiment and trust, extending the classical SIR model to depict the interaction of positive and negative contagions. The network consists of nodes, where each node has two intrinsic qualities: sentiment polarity and trust score (calculated using degree and PageRank centrality), which indicate emotional influence and credibility, respectively. The results from extensive experiments on both artificially generated data and real-world data from Facebook show that high-trust users not only speed up the dissemination process but also secure the sentiment of the entire network, whereas negative contagions mostly prevail in low-trust areas. The Graph Convolutional Network EvolveGCN, which learns the dynamics directly from the sequential graph states, was used to capture the temporal progression and structural dependencies of such diffusion. The model is very precise in predicting the final emotional states of users, achieving more than 70% accuracy in reproducing the actual diffusion outcomes. This study emphasizes that the combination of sentiment, trust, and learning through graphs results in a data-driven explanation of the interrelation between information, emotion, credibility, and social contagion.