First and foremost thank you for taking the time to read this .
I am currently running into issues when testing with extracting entities. In this case I have an intent called problem. The purpose of this is to identify if users are having IT issues mainly so I have a bunch of different training cases such as:
(I added the spaces after the [word] because otherwise it hyperlinks the word)
- I need help with my [keyboard] (problem_object)
- I have an issue with my [keyboard] (problem_object)
- my [mouse] (problem_object) is broken
- I have this [keyboard] (problem_object) that stopped working
You get the idea. Now the issue becomes when I am testing extracting the ‘problem_object’ entity. Lets say I type ‘I have this mouse that stopped working’. It will not find the entity for some reason. For whatever reason if I add another instance of something broken with a mouse in my training data (brings it up to 6 mentions of the word ‘mouse’) it now will find the mouse entity. I use the word email twice yet it is able to find that if I say something like ‘I have this email that stopped working’ (I know this doesn’t make sense but I was just testing with it.
Does anyone have any idea why this may be the case? Do I really need to reword the same sentence with every single ‘object’ I can think of?
I am using the supervised_embeddings pipeline but I also find this issue when using the one with spacy.
I literally just started all of this today so I’m sorry if this is a stupid question