Extract entity values based on words *around* an entity


I have a list of around 300 entities of varied length (some 1 word, up to about 7) and I want to pick them out. They vary considerably in content e.g. videogame names.

For example, my intents look like this:

Tell me about the game <game>
I want to know about <game> please
Search for <game>
Give me <game>
show me <game>

I want RASA to extract what it thinks is the entity based on the position of the entity in the input, not the content of the entity itself.

What is the best way to do this? If I remember, the Stanford Classifier worked in a similar way to this (but it’s been a while since I used it!).

Can I get any recommendations or thoughts? DIET and CRF don’t seem to work like this (although I am not an expert in configuring these - yet).


1 Like