Classic diffusion models assume content spreads passively — you see a post because a friend shared it. But many real campaigns in private social networks (WeChat-style ecosystems, online games) spread through invitations: a user actively invites specific friends, who must accept to join. That changes the physics of diffusion.

The ICI model

The Invitation-aware Cascade with Incentives (ICI) model captures this two-step mechanic: an invite decision by the sender, then an accept decision by the receiver, each with its own probability structure and incentive sensitivity. The model is learned from large-scale invitation logs from Tencent games and reproduces observed cascade shapes far better than passive-diffusion baselines.

Why it matters

With a faithful diffusion model, campaign designers can answer operational questions: who to seed, how to size incentives, and how invitation quotas shape reach. This work was done jointly with Tencent's Common Data Platform; a Chinese-language introduction is available on the 腾讯云开发者 blog (see the WeChat link above).