Unsupervised anomaly detection in heterogeneous graphs?

[monk1337]

Hi, I am trying to find anomalies in heterogeneous or Knowledge graphs. For example, if I use the famous MovieLens dataset and construct a heterogeneous graph, How do I go about finding anomalies as no ground truth or labels are available?

Guidance with examples would be really helpful.
Thank you!

[rusty1s]

I am not really knowledge about anomaly detection, but there exists an interesting library built upon PyG you might take inspiration from: https://github.com/pygod-team/pygod

Other than that, we provide an example of unsupervised learning on heterogeneous graphs in examples/hetero/dmgi_unsup.py which you might be able to leverage to find anomalies in output embeddings.