A context-based communication system enables the indirect addressing and routing of messages according to the users' contexts. However, for a targeted forwarding of messages towards users, the routers require a certain amount of state information. Global knowledge, i.e. each router knowing about every user, is not scalable, though, because of the necessary update messages to keep this information up-to-date.

To address this challenge, a router can aggregate similar contexts and only provide such an aggregated view to neighboring routers. However, due to different requirements and semantics, existing approaches from publish/subscribe or data clustering are not applicable in a context-based communication system. We therefore present an approach to aggregate two similar contexts, based on a similarity measure for user contexts. The algorithm can be adjusted according to the observed messages and user contexts in the system by specifying a similarity threshold to determine when contexts are aggregated.

We also present a qualitative analysis of the behavior of the system under different load scenarios and their respective appropriate similarity thresholds.