Error in Output Alignment Between Citation and Dataset

(0) Share

Report

Posted on by LM-12011854-0

I’m encountering an issue where my Microsoft Copilot bot does not produce accurate or aligned output, particularly when generating citations based on my dataset. Despite providing clear instructions, the bot often:

Mismatched URLs and IDs

The bot occasionally provides a URL that does not match the ID for a given resource.
This mismatch leads to user confusion and undermines the accuracy of the recommendations.

Omitted or Misplaced Fields:

Occasionally, fields from the citation are sourced from different resources in the dataset, causing incorrect or mixed data in the output.

Parsing Issues with Dataset:

It appears that the bot does not fully understand how my dataset is formatted or fails during parsing.
This often results in citations that include partial or unrelated information from a different ID, typically from the previous or next entry in the dataset.

Environment Details:

Using Microsoft Copilot in [insert your development environment].
Dataset is structured in a flat-file format (TXT). I uploaded my file as knowledge.

Request for Support:

Could someone guide me on how to configure the bot or refine my instructions to prevent mismatches and ensure complete, accurate citations?
Are there specific ways to improve how Copilot handles structured data outputs like this?
Request for Support:
How can I configure Copilot to handle structured datasets without mixing fields?
Are there specific parsing techniques or validation rules that can prevent these mismatches?
Can Copilot be trained to better understand sequential datasets and ensure accurate mapping?

Categories:

General topics

I have the same question (0)

All responses (2)

Answers (0)

Suggested answer

Artur Stepniak 1,539 Moderator on at

Like (0)

Report
Copy link

Link copied!

Hello,

how this TXT file is formatted? Is it tab-separated maybe? I think you should consider switchting the datasource. You could reformat the file to CSV, for example or use a Dataverse table. It seems that it indexed the file as a text file, thus it just gives you a response based on the index search, it's not parsing row by row.

In case of any other questions, let me know. If the answer helped you, mark it, so that others can benefit from it.

Best regards,

Artur Stepniak

Interested in GenAI? Visit my site!

Was this reply helpful? Yes No
Suggested answer

LM-12011854-0 11 on at

Like (0)

Report
Copy link

Link copied!
Hi Artur,

Thanks for your prompt response.

Here are my answers:

How is this TXT file formatted? It is tab-separated.

The solution was working pretty well, but now it is responding with a high error rate because of this mismatch.

I have tried CSV, but the responses are not as relevant as with the text file.

I tried Dataverse, but my team wants to stick with the text file because of the tokenization.

I would appreciate your response if you have any different ideas. Please let me know.

I'm not sure if we could do anything else, as the dataset is supposed to be parsed when I add it as a knowledge source. That is my understanding.

Regards,

Was this reply helpful? Yes No