Skip to main content

Notifications

Community site session details

Community site session details

Session Id : AaCpYiUqgAudf12YlG1FSN
Copilot Studio - General
Unanswered

HR-Chatbot

Like (0) ShareShare
ReportReport
Posted on 5 Feb 2024 13:57:51 by 16

Am developing a chatbot for HR. I've uploaded som dokuments to make the chatbot use generative answers function to answe based on the dokuments that i've uploaded. Sometimes answers the questions perfectly but sometimes he gives wrong answers even thogh it's the same question. Can someone explain why it happens? How can i make him smarter to answer the questions and understanding them?

Categories:
  • reveriaa Profile Picture
    2 on 13 May 2024 at 07:33:03
    Re: HR-Chatbot

    To make your chatbot smarter and more effective in understanding and answering questions:

    1. Fine-tune the Model: Continuously train and fine-tune your chatbot using additional data, including feedback from user interactions. This can help improve its ability to understand and respond to queries accurately.

    2. Implement NLP Techniques: Utilize Natural Language Processing (NLP) techniques such as named entity recognition, sentiment analysis, and context modeling to enhance the chatbot's understanding of user queries and context.

    3. Feedback Loop: Implement a feedback loop mechanism where users can provide feedback on the accuracy and relevance of the chatbot's responses. This feedback can be used to iteratively improve the model over time.

    4. Regular Updates: Keep updating the chatbot with new documents, data, and improvements to ensure it stays relevant and up-to-date with the latest information and trends in HR.

  • AAschauer Profile Picture
    Microsoft Employee on 01 Mar 2024 at 17:14:26
    Re: HR-Chatbot

    @Ameeralkarmo - keep in mind what you are using is a Large Language Model. It basically takes a text input and computes the most probably completion - so there is no concept of "understanding".

    You have to be mindful that changing a question in the slightest (adding "please" for example) might change the output (as it is determined as a probability). 
    What you can try is playing with Custom Instructions. These instructions become part of the models prompt and influence amongst others the tone and format of the responses:
    Enhance your copilot's responses with prompt modification - Microsoft Copilot Studio | Microsoft Learn

    Also take a look at this guide for prompting (which is applicable to the GPT family of LLMs) to help you with possible instructions to the model:
    Prompt engineering - OpenAI API

Under review

Thank you for your reply! To ensure a great experience for everyone, your content is awaiting approval by our Community Managers. Please check back later.

Helpful resources

Quick Links

🌸 Community Spring Festival 2025 Challenge Winners! 🌸

Congratulations to all our community participants!

Warren Belz – Community Spotlight

We are honored to recognize Warren Belz as our May 2025 Community…

Congratulations to the April Top 10 Community Stars!

Thanks for all your good work in the Community!

Leaderboard > Copilot Studio - General

#1
Romain The Low-Code Bearded Bear Profile Picture

Romain The Low-Code... 23

#2
Pablo Roldan Profile Picture

Pablo Roldan 16

#3
stampcoin Profile Picture

stampcoin 10

Overall leaderboard
Loading started