0

We are building a chatbot application for our ERP Project.

In our system we are having more than 500 FAQ's questions (We Can't group questions, each QA is different), how do I create intents for all questions because AWS Lex is having the limitation of 250 Intents.

If I use Lambda, we should build the NLU. Anybody suggest me what is the best approach to build the chatbot in the above scenario.

John Rotenstein
  • 241,921
  • 22
  • 380
  • 470
Sathishkumar
  • 3,394
  • 4
  • 20
  • 23
  • 2
    You seem to already know what you need to do. Group the questions into intents as best you can and use slots within those intents to narrow down the exact question. Validate in Lambda really well to ensure you are understanding exactly what the user is looking for. Having 1 or 2 back and forth clarification questions (more slots) won't hurt. Then build the answers based on slots, not based on intents. If you are sure you can't group them, then share a good 10-20 of them and I'll see if I can surprise you =) – Jay A. Little Jun 26 '20 at 11:55
  • 1
    Also, from my own experience with FAQ handling in Lex, it is better to generalize the answers and provide more information than what is being asked in order to satisfy the variety of specific questions that can be handled with one complete answer. – Jay A. Little Jun 26 '20 at 11:58
  • I would concur with @jay-a-little on using the slot values to build the appropriate response. Trying the answer-per-intent approach is easy and works at first but you will run into these scalability issues once the scope of the Bot starts growing. You'll also find that it gets difficult to fashion the utterances as the topics start getting a lot closer - this is a good sign that you need to bring in more slots with clarifying prompts and then letting lambda do some heavy lifting for you in terms of returning the correct response. – Reegz Jul 23 '20 at 22:26

0 Answers0