Just start by adding a list of several annotated sentences to your intent, and include
Something like this, where I named the
entity to extract
- I want to search document start with [technical](topic)
- I want to search document start with [human](topic)
- I want to search document start with [classes](topic)
- I want to search document start with [ability](topic)
- I want to search document start with [locations](topic)
- I want to search document start with [integration](topic)
A few things to do as well:
- You must add
topic as an entity to your
- It will be very good to use entity synonyms. For example, if some one uses
integrate instead of
integration, the entity extracted will always be
- You must make sure that the
CRFEntityExtractor is used.
If you train the model, it will be able to extract the entity
topic for the sentences given, and it will generalize so it will also pick up if the user spells the topics a little different.
Now, your users will never ask it exactly like this. They might ask
Look up documents with technical info. So, you need to make sure that you have a lot of possible sentences in your training data, and that they are all correctly annotated.
It is impossible for you to come up with all the different ways your users will ask things. That is why you should first build a simple assistant to correctly handle some paths, using Rasa Open Source. Once that is kind of working, let others talk to your bot to collect more data, using Rasa X. Here is a nice blog post about that process.