16
Chapter 2. The core model
an equal chance of transitioning to any one of the other layers.
As we shall see, there are settings where it makes sense to adjust the relative
weighting of within-layer and between-layer edges, but the motivation in terms of
random walks provides a principled starting point from which to motivate deviations
when they are needed.
2.3 Summary
The key construction that we will use to capture rich edge semantics is to replicate
each node of the social network into multiple versions in the graph, connect edges
to the appropriate version(s) to capture their semantics, and add edges between the
multiple versions as necessary to keep them aligned. The resulting graph is notionally larger, but the additions are only linear in size so that the representation and
computations also grow only linearly. The larger graph is embedded in a more or
less standard way, but the resulting embedded graph has nodes and edges of different
kinds, and so the downstream analysis changes because there are many more possible
structures to understand and exploit.