- GCN - PinSage which combines efficient random walks and graph convolutions to generate embeddings of user-item pairs
- Embeddings incorportage graph structure as well as node feature information
- Trained using pairwaise ranking loss function and able to handle sparsity by using negative sampling.
- Negative sampling is used to generate negative examples for the pairwise ranking loss function.
- These examples are generated by randomly sampling items that are not connected to the user in the graph structure of the data.
- Pairwise ranking loss function:
- 2 parts - positive: loss for the correct ordering of the items
- 2 parts - negative: loss for the incorrect ordering of the items
- Total loss: sum of both losses
- Trained on 3 billion nodes representing pins and boards and 18 billion edges - according to A/B tests, PinSage generates higher quality reccomendations than other DL alternatives.
- Importance pooling: feature information is aggregated from local neighborhoods in the graph - but we introduce a method to weigh the importance of node features, based upon random-walk similarty measures.
Loss: 