I am looking to build n entity relation extraction model. While creating the training data I have a confusion. Check the following example:
"I live in India. My brother stays in USA."
Say I want to extract relation: "Lives In". Is it enough to tag "I" --> "India", "brother"-->"USA" while creating the training data.
Or do I need to have another relation like "Not Lives In" and create training data marking the relation between : "I" -->"USA and "brother"-->"India".
If you're annotating with Prodigy, you only need to annotate this part (not the negative relation). Watch this part of Sofie's video where she explains:
For more details on the internals of relation extraction, check out the spaCy documentation on it.
Related, the Prodigy docs have the data schema for Prodigy's
relation UI annotations.
Let us know if you have further questions!