How can I train a bot to answer FAQ which are quiet length?

We are having well documented FAQ for a software product. FAQ aren’t that simple, having very lengthy answers. our company wants to build a bot so that it should be able to answer those question.

How can I create intent & entities for such complex data. Please suggest some good approach

I understand your concern i also had to go through this. You can do one thing see you will have to go through the pain for training the bot with intents and data as it needs to understand what user will input.

You can either work with RASA’s NLU format itself and for that tensorflow pipeline will be the best approach for that.

Now in your UI in which ever language you are using , you can save these responses in a database, config file like xml or etc . So that you just have to query the bot for the intent name only and you can match that intent with the file or db , in this way you can query heavy responses very easily!

1 Like

Thanks for suggesting. This could be one of the approach but isn’t an optimal approach

it is infact one of the most optimal solution for FAQ in which user query’s a question and bot gets the answer but if you want to have a conversation which you didn’t mentioned then you will have to go through implementing rasa core for that!

how should we implement using rasa_core ? how we could use for this purpose ?

I’m also new to this , but i have the same scenario .

I think it comes down to taking unstructured data( text) and converting it to intents and entities.

I’m starting with extracting the entities using spacy or scikit learn. I’ve just come to realise that the data we have to work with is just wrong. Some paragraphs will work some wont.

it does seem to me that the training data or NLU online trainer has some good scenarios but the questions are less specific than that of an direct q/a relationship table.