I was wondering if Rasa’s entity detection is on an intent-to-intent basis. I have a nice set of generalized sample data in one of my intents for recognizing company names, but it seems as if only that intent will take advantage of the labeled entities and not any other intents that utilize the same entity in its phrases. Is this how it was designed or am I making false assumptions? I am using supervised embeddings and CRF Entity Extractor
Hi @bhargavat, Rasa entities are not intent specific, so they can be used across intents. However, for entities to be detected correctly they need to always be labeled appropriately (even across different intents). Hope this helps!
Hi @fede. Yes I am aware of the entities needing labeling everywhere. My question is say I have 50 entities in my lookup table. If I have 3 intents that share this entity group, could I train these 50 across these intents and have all 50 of them be detected equally across these 3 intents? Would there be a higher likeliness of detecting an entity from the list of 50 for a particular intent if it was used in one of its training phrases? I feel like I am seeing this type of behavior. An entity from my lookup table gets detected within the phrases of an intent where it was trained, but failed to be detected in phrases outside of the intent in which it was trained under. I believe this is a random 50/50 chance that this happens. There is no guarantee if it will be detected outside of the intent it was trained under, but it’s still possible that it does.