[arangodb-google] Pregel friendly graph structure

[Victor Genin]

We want to represent social engagement inside the graph (userX followed
userY on 2012-12-23, userX followed userZ on 2012-12-25, where follow might
be any kind of engagement like commenting, watching video...)

The question that we would like to ask is if userX follows userY, will it
lead to following userZ. To solve this we would need to take edges starting
ending at userY (all followers of userY) and see who of them follow userZ
looking at the date difference distribution.

We were thinking about 2 approaches, one with dates of engagements are
represented as nodes in the graph and the other as edge attributes (see
attached). The idea behind dates as nodes was to look for cycles and count
the number of traversals, which essentially are the number of days between
user following userY and userZ.

The approach with dates on edges looks "regular" and database friendly, but
poses questions how to implement similar algorithms.

To summarize I guess, my question is whether putting more logic onto the
edges limit the possibilities of using graph theory algorithms (Jakarta,
graph cut, min span) and should one aspire to put as little logic on the
edges as possible to achieve this.