GraphRAG visualization and UI

[natoverse]

Placeholder discussion for visualization and UI, which community members have commonly posted examples and solutions for

[natoverse]

#566

[natoverse]

#689

[natoverse]

#427

[natoverse]

#576

[natoverse]

Some notes I've put elsewhere:
A couple of suggestions:

1. When running your index, you can set GRAPHRAG_SNAPSHOT_GRAPHML=True to output graphml files that can be visualized in a desktop tool like Gephi.
2. We have config for node2vec and umap, which will embed the graph and then do dimensionality reduction so each node is assigned an x/y coordinate in the output. This is useful for plotting programs, and will locate the nodes near their community clusters.

[OMGLASERSPEWPEWPEW]

Sorry I'm a newbie developer. When I look at the settings.yaml, I can't find that setting. Is it somewhere else? I installed graphrag via pip install and it's the latest version 0,2,2

[BobbyAxelrods]

Sorry I'm a newbie developer. When I look at the settings.yaml, I can't find that setting. Is it somewhere else? I installed graphrag via pip install and it's the latest version 0,2,2

Its there,

1. Enable the snapshot of graphml = true & umap = true . Good to go
   

Other option you may upload artificat to viz the network
#884

Hope this helps

[KylinMountain]

@natoverse so when enable the umap and node2vec，what is the output file name? And how can generate a 3D graphml? Any suggestion is appreciated thank you.

[natoverse]

The create_final_nodes.parquet has an x and y for each node. These are set to 0 by default, but if you enable umap and node2vec they will be populated with a coordinate that you can plot. We don't currently expose the ability to compute a z position for 3D plotting.

[KylinMountain]

besides i have another question about is there any weight to metric the entity or relationship importance? Thansk

[natoverse]

The rank field for entity nodes matches the degree - i.e., we simply use the node degree as the rank. For the relationship rank we sum the degrees of the two connected nodes. So both of those consider the connectedness of the nodes for assigning a weight.

[KylinMountain]

thank you very much, this is very helpful to me. What rank the really meanings of the entity? Looks like we didn’t add count field to the entity, is this also important in graph knowledge?

[natoverse]

The rank is the count of connected edges, which can be computed from the edge list without storing any count on the entity.