Response retrieval model allow simple QA scenarios. The problem seems to be from what I tested so far two fold.
First the annoying one … you have to sync the editing of 4 different files … it is really hard keep track. It would be probably better to have single file (probably different extension that hold all the information. This way also no need to repeat declarations.
Second … it seem that it support one-to-one OR many-to-one relations i.e.
Question ==> Answer Questions ==> Answer
but not :
Question <== Answers Questions <== Answers
i.e. probably a training on all permutations of all Q <=> A pairs can do it.
q1,q2 <=> a1,a2
q1,a1 q2,a1 q1,a2 q2,a2
What I’m saying is the current structure support MAPPING from multiple INTENTS/QUESTIONS to single ACTION/RESPONSE, but not from multiple ACTIONS/RESPONSES to single INTENT/QUESTION.
I wrote this quick script to generate the files from single file : qa2md.py (1.0 KB) )
Example item :
==== # variables, data_types * what are the standard data types ? * give me list of Python supported types - numbers, string, list, tuple, dictionary