Traffic models for community-based ranking and navigation

We investigate multinomial mixture traffic models for community based ranking and navigation. A "highway" model of source-destination traffic is formulated which aggregates the traffic through an underlying network of highways, onramps and offramps. The model extracts community structure from source-destination traffic information, but in addition captures the aggregate "highway" traffic between the communities. This important distinction extends the highway traffic analysis beyond clustering. The analysis discovers communities, rankings of the communities, rankings of destinations within each community, and transition probabilities between the communities. The model can be used for community-based collaborative filtering when applied to similarity graphs for music or movies.