3

I am looking at creating a simple chatbot which can use a pdf file as it's source. For example, the input to the chatbot can be a bank's terms and conditions document and the chatbot would respond to a question which are related to the contents of the document.

Sample Q&A. Q : What is my monthly fee for my savings account? A : Your monthly fees is $5 for the savings account if no deposit is made above $2000, else free.

I used pdfminer to read the pdf document and convert it into processed data and spaCy to identify the NER, POS etc.

I learnt about RASA and all the links which I saw uses a defined text response and not using any pdf document as a source.

Can someone provide any guidance on which approach i could follow?

I don't want to use Dialogflow or Lex and want to be in the open source world.

Ashok KS
  • 659
  • 5
  • 21
  • Its a long process. You will need to scrap and normalize those documents. You could also need to apply some information process. – tumbleweed Jun 16 '20 at 17:27

1 Answers1

1

What you are looking for is a Question Answering task which traditionally can be solved with Deep learning. In Rasa, you can mostly generate output text based on intent from a lookup table. But to find some answer based on question, you will need additional processing.

If you are looking for Question-Answer as a service, you can have a look at QnA Maker and look at this thread which also discussed other approaches.

Touhidul Alam
  • 321
  • 2
  • 11