Delete or Not: A Game-Theoretical Model for Soft Censorship of Rumor.

Rumor censorship of social media platforms has become an important issue in the academia and in practice. However, most studies focus on the complete rumor censorship behavior rather than the soft censorship behavior of (social media) platforms. To characterize soft censorship behavior, we conduct analytical, numerical, and experimental analyses using game theory to determine the specific strategies of platforms and rumormongers. Given that (1) the censorship behavior of platforms is costly and (2) platforms have a limited accuracy rate to identify rumors correctly, the platform may identify rumors as true information or identify true information as rumors; moreover, (3) rumormongers decide whether to publish rumors or not to avoid been deleted by the platforms. We found that (1) if deleting true information mistakenly has benefits rather than cost (the platform may cost less by not improving their rumor identification algorithms if the public pays less attention to the freedom of their speech), then platforms are more likely to censor rumormongers and delete the information they published; (2) if deleting true information is costly, then platforms become more cautious about their deleting behavior. This study explains why censoring is accepted by the public in some countries but is highly questionable in others. Using these findings can help platforms understand the rumor publishing behavior of rumormongers and make decisions based on certain situations.