How to handle dates?

I have a doubt regarding the dates. How to handle dates and time as they come in various forms?

In case the entity classifier fails to detect any intent(let’s say city name), is there any way to create some sort of matcher for the undetected cities?

1 Like

It’s something that I am also checking for the current Rasa version.

For some annotated and tokenized instances, I have some custom entities like DD and MM that are not captured when the model has to infer something as ‘05/10’ with DD: ‘10’ and ‘MM’: ‘05’.

@sainimohit23 @bayesianwannabe Have you tried duckling? It’s a component which is pretty good at extracting different date types

2 Likes

Thanks for your reply @Juste. I asked this because if I train the signle model for both the entity recognition and intent classification task then DIET will consider both losses in the training process. And theoritically both task will compliment each other in the training process i.e. more robust model.

Anyway I found out that just marking the dates will work.