OpenAPI Sentiment Analysis return values issue

[Mark_Eden]

Describe the bug:

• When adding Analyze Sentiment type column, some of the returned values do not match the values in the docs; I see some values as Positive or Neutral.

Expected behavior:

• The returned values (in the column) should all match the values in the docs (positive, neutral, negative), including case-sensitivity

[ThinhDinh]

Can you add some screenshots or a video of the bug? Thank you.

[Mark_Eden]

Sure thing @ThinhDinh . Hope this helps.

[Joe_Gabriele]

Is there a reason youre pointing to a prompt?

If youre looking to analyze your comments, then the Sentiment should be pointing at your comments column so it can interpret if that comment is positive, neutral or negative. You dont need a formal prompt to generate a sentiment.

[Mark_Eden]

Thanks, @Joe_Gabriele . This observation (presumably a bug) was to highlight that more than the documented 3 values are returned. Note the Positive (with uppercase “P”) and Neutral (with uppercase “N”) values in addition to the documented (all-lowercase) positive, neutral, negative values. I wouldn’t expect the input to have any influence on this.

[Joe_Gabriele]

I can confirm that there is a random neutral with a capital N.

Interestingly enough, if I change “why” to “hi”, it goes to a lowercase neutral, and if I switch it back to “why” it goes back to a Capitalized Neutral. I also tried a bunch of different words and they go to “neutral” but going back to “why” moves it to Capital again. Very strange…

[Darren_Murphy]

cc @Jeremy

[ThinhDinh]

I wonder if the new functions thing can help with this, to force OpenAI to return a more defined set of outcomes.