Slot mapping based on confidence

Rasa Open Source
1

Hi, i want to control either entity extraction or slot mapping based on the CRFEntityExtractor’s confidence score. Is there any way (other than manipulating the rasa code itself) to control that?

The issue im facing is, that a common string is mistaken as an entity value (even though its not very similar to my (100+) training examples). However it has a far lower confidence score than the ‘real’ entities.

Since I thought that should be a common problem, i was surprised I couldnt find a solution either in slot_mappings or in the CRFEnitytExtractor. Did I miss something? Or do I have to write custom code for this?

2

@IgNoRaNt23 Hi! CRFEnitytExtractor just extracts the most likely tag per word. In most cases the tag is just “others”, e.g. no entity. There is no possibility to set a specific confidence threshold. Also, you don’t have access to the confidence score in slots anymore. So, I’m afraid you have to write something yourself. However, feel free to make a feature request on GitHub.