ttm / percolation Goto Github PK

there have been years of maturing. For now, we should leave it for data to confirm and refute.
The data will give directions to deeper questions and hypotheses
and should deliver robust results within the Open Linked Social Data legacy.

It seems to only allow union graph through TDB.

the construction of uris in social, gmane and participation packages have many subtleties sorted out in code.

Of these, one remained until now: the uris of individuals for the most important entities should have the snapshotid in it or not?

Cons of using the snapshotid in the uris: the uri gets larger (and so the final data). Same entities are not directly related if they are present in different snapshots (they have to be related through IDs).

Pros of using the snapshotid in the uris: data from each snapshot does not get mixed. For example, a twitter user can have different screennames in different snapshots. The information about which snapshot each screenname is related to is lost in the final linked data.

This last difficulty can be overcome by the use of named graphs (alias "conjunctive graph contexts") and the RDF link between the instances and the snapshoturi.
Support for named graphs has some drawbacks in real usages,
such as the current inability in Fuseki/Jena for creating new named graphs on the fly and reasoning about them (the reasoner has to be loaded at startup for each graph).

The currently adopted "solution" is not to use the snapshotid in the uris.

given a fitness measure, implement GA and other algorithms to continuously find associations of measures and entities, i.e. for generic pattern recognition.
An output of the GA (probably the gene directly) might be,
for example, "messages sent between 23h-4h are more incident
from intermediary paricipants and hubs, but not the few most active hubs."
or the already found:
"the ~20 participants with greatest betweenness make almost perfectly disjunct group
with the ~20 participants with greatest closenness".
or
"Intermediary participants with high out degree use shorter words, longer words otherwise"
"the set of participants with greatest clustering and degree above 3
have local vocabulary and linguistic structure".
etc.