Power Platform Community Forum Thread Details

I am just starting to play with copilot studio and I decided to try to start with what I thought would be a simple use case. I decided to build a bot that allows a user to submit new use case ideas. I was wanting the bot to ask them first to describe their idea and then it would search in it's data source (which I have in a dataverse table) for other ideas that have a similar description. If it finds any that it would then list those out to the user and ask if these are the same as their idea or not. if they are not it would then create a record in the dataverse table with their idea.

So far I have been able to get the basic chatbot stuff working on asking the user for their idea but am having trouble with using copilot to search in its datasource. I have the table listed in the knowledge source and I have even updated the synonyms section. When I have in the topic to send the description to a generative answers step which I ask it to find similar descriptions to the following I send it is very hit and miss on actually finding anything even if I know it is the same thing. I have even tried just asking to find stuff in the generative part (no topics triggered) and it most of the time doesn't find anything. what is weird to me is if I ask it on some of the other data columns that are choices or like who submitted it (a simple text column with a name) that it finds those just fine. It seems like just the description column I have that is storing the use case it has a problem understand and searching it.

I have even just asked it basic questions of find an idea that talks about Microsoft Forms and it is not able to find any. I know there are some as I just do a basic search for Form in the column data and am able to find some.

So my questions are:

there a way to improve copilot to understand and parse this information?
is there a limitation on what type of column data copilot can parse?
am I just not thinking this is a simple use case?
Any tips or tricks on improving copilot?

Categories:

General topics

Hi erikbjornebo,

Gen AI may struggle with precise matching of free-text fields like descriptions, especially when using NLP to identify similar ideas.

Refine the synonym list further, ensuring that common alternative phrasings for descriptions are covered. You can also try guiding the bot to search more precisely by using specific terms and rephrasing the user’s input in a format the AI better understands.

Implement a fallback mechanism. If the AI fails to find relevant descriptions, you can create a fallback logic that either asks the user for more details or triggers a more refined search flow.

Text fields those with unstructured data (like a description of ideas), can be trickier for Copilot to parse accurately. Columns that include rich text, HTML also may not be interpreted well unless parsed properly by Copilot.

Collecting a description from the user, searching for similar descriptions in the database looks simple. But complexity arises when dealing with free-text field.

For more advanced natural language understanding, consider using AI Builder alongside Copilot. This will allow you to perform sentiment analysis, category extraction, and other AI-driven processing that improves the interpretation of unstructured text fields.

For complex searches try to use Power Automate workflows with Copilot. You can create a flow to handle more complex logic (like similarity search) and return the result to Copilot.

(I have'nt tried this)

For longer or more complex descriptions, apply natural language processing (NLP) techniques (such as semantic search) using Azure Cognitive Services or similar tools to better match ideas.

Thanks.