Closed jesweare closed 1 year ago
@jesweare Thanks for your feedback! We will investigate and update as appropriate.
@jesweare
Thanks for your feedback! We have assigned the issue to author and will provide further updates.
Hi @jesweare, the team is aware of this issue and is looking into it. Thank you for reporting the behavior!
In the Notes section, there is a passage below:
that implies synonym orders would not impact the results of the prediction process.
However, that is not the case; changing the synonym order may create a difference (albeit trivial) in the
confidenceScore
value, given the Q&A pairs, synonyms, and query are all identical.The Product Team has confirmed that the score difference is expected, as "computational logic" is representative of only a portion of the whole prediction and ranking process. I assume this rationale can be inferred from the old QnA Maker documentation/diagram explaining the ranker process: How QnA Maker processes a user query to select the best answer
If that is the case, the documentation should add points that makes it clear that score fluctuation may happen. Example below.
Document Details
⚠ Do not edit this section. It is required for learn.microsoft.com ➟ GitHub issue linking.