Here is a link to the slides (Google slides)

Presented at Analytics Forward 2023, Raleigh, NC, on 3/18/2023; hosted by Research Triangle Analysts (meetup group)

Abstract:

Graph Convolutional Networks (GCNs) are a powerful addition to the machine learning toolbox for classification problems. GCNs use the adajency matrix representation of a local subgraph to take the features of nearby nodes into account when predicting the class of the center node. This approach has been reported to greatly increase recall on classification problems, such as reCAPTCHA offline BOT and HUMAN labeling tasks, because suspicious BOT nodes can be identified more readily in clusters than individually.