AWS Marketplace - Machine Learning solutions

ML (Machine Learning) can be subject to political ambitions. You can tune library the way, it gives certain predefined results, based on developer's/company's political views.

When we are analyzing several solutions from AWS Marketplace sections Machine Learning - Text we have figured out that some solutions are politically oriented.

Here is the test:

1. We use CNN webpage to pull article - please, note link is valid as of 09/13/20
2. We used paragraph 3

"Trump will visit McClellan Park, California, on Monday for a briefing with local and federal fire and emergency officials on the state's wildfires, the White House said Saturday. The President was already set to visit Reno, Nevada, on Saturday, where he is likely to see firsthand evidence of the devastating fires, as the National Weather Service in Reno issued a dense smoke advisory through Sunday"

1. Tested Sentiments in 3 different solutions
2. Twinword API
3. AWS Comprehend

4. SigmoData Sentiment Analysis

5. Method used - Replace word Trump with word Biden

6. The results:

   • Twinword API - same scores
   • AWS Comprehend - same score
   • SigmoData Sentiment Analysis came with surprise
     • With word Trump it gives 38% of positive score
     • With word Biden it gives 82% of positive score Try it yourself https://aws.amazon.com/marketplace/ai/model-evaluation?productId=702afdb5-f282-4579-9f4d-d37728f74bf5

Summary: before using any ML model, make sure you test it on edge values.