use case · wild
Run NER over every sentence in a paragraph and something bigger falls out: a graph. Each entity becomes a node; two entities are connected when they're mentioned in the same sentence. Feed it a news story and you get a live map of who's involved with what — the raw material of knowledge graphs, built entirely on your device.
Edges here are simple sentence co-occurrence — the same entity written two ways ("Angela Merkel" / "Merkel") shows as two nodes, and the graph doesn't know the type of relationship. That's where a second model comes in: the multi-model demo uses embeddings to link mentions that mean the same thing.
← Practical · Model · Next: Multi-model →