The construction of "embeddingv_1" variables in the Google Patents database

Google Cloud Database
1

I would like to confirm which information was utilized to construct the embedding_v1 variable, which is stored in the google_patents_research.publications table. Specifically, was the model trained using:

  1. The full patent text (specifications)?

  2. Images (patent drawings)?

  3. Citation data (references)?

Were all three of these information sources used?"