In many campaigns, an influencer's attention is a bounded resource: a stream host can interact with only so many viewers, and an event organizer can serve only so many participants. Classic influence maximization ignores this, assuming a seed influences unboundedly. Capacity Constrained IM (CIM) models each seed with a capacity on how many activations it can actually serve.

The technical challenge

Capacity turns seed selection into a joint selection-and-assignment problem: choosing seeds and deciding which influenced users each seed serves. The objective loses the clean submodularity that classic greedy exploits, so this work develops new algorithms with provable guarantees and scalable implementations based on reverse-reachable sampling.

In practice

Experiments on large social networks — including production scenarios at Tencent — show the capacity-aware algorithms deliver substantially more served users than capacity-oblivious baselines under identical budgets. A Chinese-language 腾讯云开发者 article on the "super spreaders" behind this line of work is linked above.