Azure text analytics accuracy in german is different from the demo case

Question

I've created a website which communicates with my logic app while that uses the text analytics from azure. But my applications acts different from the demo case which you can find here: https://azure.microsoft.com/de-de/services/cognitive-services/text-analytics/ ! When you post for example:

Hallo E-Bike Team,

ich habe ein Problem mit meinem E-Bike. Es ist das Model ProRide2E. Seit heute Morgen steht auf dem Display folgender Hinweis:

„Akkuleistung beeinträchtigt. Fehlercode: XB1200AB“

Das darf doch wohl nicht wahr sein. Ich bin echt sauer. Wieso ist das doofe Bike immer kaputt?

Ein genervter Kunde

Name

which is very negative it responses with a sentimental score of 31% in the demo but in my app it responses with 50% which is clearly wrong because it is negative and it should be below 50%. I uses the same cognitive services as the demo but my accuracy is not similar to the demo.

Is there any way to improve my accuracy?

ps: I'm using the free subscription. Does the accuracy change if change that?

Answer 1

I tried to repro the issue at my side and found that when you don't set Language parameter it will give you the SENTIMENT score of 0.5

When you set the Language parameter to de it will give you exact SENTIMENT score of 0.30392158031463623.

Below is the screenshot to show how you can set the Language parameter: