Cortex AI logo placeholderCortexLabAI · learning · engineering
Model Atlas · Graph learning

Graph Neural Network

Learns node/edge representations by passing messages across graph neighborhoods.

core mathematical viewhᵥ′=UPDATE(hᵥ,AGG({hᵤ:u∈N(v)}))
Mental model

Understand it before memorizing it.

Each node updates itself using messages from its neighbors.

Best fit

Where this model earns its place

Knowledge graphs

Start with a simpler baseline then compare this model using the same evaluation split and operational constraints.

Fraud networks

Start with a simpler baseline then compare this model using the same evaluation split and operational constraints.

Recommendations

Start with a simpler baseline then compare this model using the same evaluation split and operational constraints.

Strengths and limits

Trade-offs matter more than popularity.

Strengths

✓ Uses relationships directly

✓ Flexible graph structure

Limitations

△ Oversmoothing

△ Sampling complexity

△ Hard debugging

Evaluation

Metrics to watch

Hits@kInterpret with the product objective and error cost.
MRRInterpret with the product objective and error cost.
Node F1Interpret with the product objective and error cost.
LatencyInterpret with the product objective and error cost.
Production checklist

Before it reaches real users

  1. 01

    Control neighborhood sampling

    Document the assumption and instrument the condition so regressions can be detected.

  2. 02

    Watch leakage across graph

    Document the assumption and instrument the condition so regressions can be detected.

  3. 03

    Evaluate cold-start nodes

    Document the assumption and instrument the condition so regressions can be detected.