How to do good information retrieval in practice? Experience wanted

Rasa Open Source
1

Hey, the development of the chatbot is just one key point on your to do list in general. Right now, I struggle how to read the DB/Knowledge Base or just the json file properly to extract the right answer. In my use case I have very much Keyword names/entities together with synonyms. So it would be a bad idea to write the synonyms in the NLU training file manually. My file has first listed the keyword and afterwards the answers are separated for each intent belonging to this entity. How do you handle in practice such keyword search together with slightly different keywords/synonyms given by the user?

I am interested in standard approach to this and in future I would like to apply NLP maybe like word emebedding for my custom case.

2

do you mean extracting a list of predefined values for an entity? have you looked into using a lookup table? http://rasa.com/docs/nlu/master/dataformat/#lookup-tables

3

This has nothing to do with Rasa. Just wanted experience on how to do a good search, when you have keyword which is not in the data but maybe a synonym or other spelling. I suppose the only way to go is to create a dict with all domain keyords and cluster them to topics. One other way might be word embeddings or just list synonyms in NLU training file…