Watson Discovery

 View Only
  • 1.  Watson Assistant Integration

    Posted Fri January 27, 2023 10:25 AM
    Hi, I'm trying to make an assistant and integrate it with Watson Discovery. I'm using Bahasa Indonesia as the language for both the assistant and the document in discovery. I used the user-defined model options and splits the document by it's subtitle. My goal is that when the chatbot user asks something related to a subtitle, the assistant return the related text. When I try to ask with the exact sentence as the subtitle in the document, there is no problem. But when I ask the same intent with a different sentence, the assistant can't find the answer. For example, in the document there is a subtitle "Definition of XYZ", but when I try to ask "What is XYZ?" the assistant can't find the answer. I don't know if it is related with the language that I used. I want to know what is the problem here and is there any possible way to overcome this problem?

    #Watson Assistant #Watson Discovery


  • 2.  RE: Watson Assistant Integration

    Posted Mon February 13, 2023 04:18 PM
    Edited by Antonel Neculai Mon February 13, 2023 04:29 PM

    Irfan, there is an option  in Discovery to train it by selecting the questions and indicating which answer is the correct one. You can select the question "What is XYZ?"  pushed  by the assistant into Discovery and select the correct subtitle "Definition of XYZ". I assume Discovery will learn to associate the two ways of asking the question about XYZ and become better in time at identifying the answer. I guess after a few variations of the same question and you indicating for it the correct answer, it will smarten up enough to be able to carry the task by itself.

    Here is documentation about training Discovery for improved result relevance:


    Antonel Neculai

  • 3.  RE: Watson Assistant Integration

    Posted Mon February 13, 2023 10:46 PM

    Hi, Thank you. I ended up using intent in Watson Assistant to list the possible ways to ask the question. I think the result is good enough but I found a new issue.  Somehow the Watson Assistant didn't give a complete response that Watson Discovery gave. Here is a photo of the response on Watson Assistant which is incomplete and the ideal response as shown in the Watson Discovery. Do you happen to know why it happen?

    Irfan Setyanto