Hey everyone.
I'm new to Copilot studio and I had a question. I'm trying to create a bot that is based on an internal document of our company. I was able to upload the document and publish the bot, but when I started testing it I found that it gives back inaccurate answers. On further investigation of the citations given with the answers, I found that most of the issues arise from the fact that the document is full of tables.
My question is, do these documents that are used as data sources need to be exclusively text or paragraphs of text? or is there a way to configure the bot to use the data in the PDF/Word tables to give back accurate answers? I'm mainly asking if the convention for chatbots to work correctly is to make the document text only.
I would really appreciate any feedback or guidance on this matter. Thanks.
Hi,
Tables can be tricky, especially if they are complex with a lot of columns and lines. Try to simplify the format of your document if possible and get some of the data out of the table. You can also try to completely convert your pdf / word to a text file and upload it to your copilot, I've seen good results doing that.
You have to understand that your document is transformed to basic text when sent to the Large Language Model and that tables are lost in the process.
WarrenBelz
146,651
Most Valuable Professional
RandyHayes
76,287
Super User 2024 Season 1
Pstork1
65,999
Most Valuable Professional