2

I am working on a bot that uses sentiment analysis to analyze questions from people in the health care insurance realm. One of the main things we do is connect people with care through a find care tool. The basic questions people ask are in the form "I need a new doctor" or "I need to see a dermatologist". When the sentiment is negative we want to pass the user on to customer service, but as we just discovered, any statement starting with "I need" is considered a negative statement by LUIS' built in sentiment analysis. This is a big problem for us!

If there is a way to tweak that, please someone let me know, though I don't hold out much hope. But I would like to understand - why was that decision made? Why is a simple "I need" question considered negative instead of neutral?

hengist
  • 105
  • 1
  • 9
  • 1
    Please see my older question https://stackoverflow.com/questions/48428284/what-can-i-do-with-inconsistent-sentiment-detection-provided-by-cognitive-servic Unfortunately the default sentiment detection in LUIS has the same base. – Ferdinand Fejskid Nov 08 '18 at 11:42
  • thank you @FerdinandFejskid I will attempt to contact the machine learning team directly at the email in your answer and see if there is anything I can do. Btw, I did not find your original question when I searched, so thanks again. – hengist Nov 08 '18 at 18:36

1 Answers1

1

Sentiment analysis cannot be changed as LUIS relies on Microsoft's Text Analytics to produce a result. In addition to Ferdinand's comment suggestion, you could do one of the following, as well.

Suggest different utterances that produce a more positive result. Using 'find a doctor' produces a positive result where as 'I need a doctor' does not.

The Azure team has a ML learning lab repo that shows how to develop your own text analytics. It would require more effort but would offer you the ability to get the kind of sentiment analysis you are looking for.

Steven Kanberg
  • 6,078
  • 2
  • 16
  • 35