Color-code edges by number of co-occurrences in the co-occurrences graph?

[albertocottica]

Today I am trying to find "novelty" in the opencare data. I chose migration as a starting point. That turns out to have many connections:

I have two problems.

1. An interaction problem. Edges are too close to each other and it is difficult to select one. I can zoom, but then I lose sight of the source/target.
2. A data problem. Since we are interested in collective intelligence, I care mostly about edges that encode more than one co-occurrence. These edges are quite rare. The vast majority only encode one.

In order to zero in on the relevant edges, we could (a) color-code edges by number of co-occurrences; or (b) adopt a filter, same as in the full co-occurrences graph.

[jason-vallet]

I have opted for (b). I feared the colour coding would not have been very efficient on denser graphs.

[albertocottica]

The filter does not work well. For example, the edge between legality and safety does not get included in the graph until I crank the filter almost all the way down, and yet it encodes 7 co-occurrences. The edge between legality and migration only encodes 6, but it stays in after the former one has been filtered out.

[jason-vallet]

I had a bug yesterday evening when trying to fix #15. Otherwise, the filter's progression is linear so there is some gaps for which the filter will not have any effect. Tags with numerous co-occurences are responsible for corrupting the scale.

I have replayed your example but I have different results. The edge connecting legalityand safety weights 7 co-occurrences but the one between legality and migration is much heavier, encoding 21 co-occurrences which would explain the behaviour.

You can visualise the edges' label to check the number of co-occurrences being encoded by the edge.