Improving the question answering model

It is very important to monitor and continuously improve a machine learning model. The same goes for the model we created when using the question answering service. Even though it works perfectly when we as designers test it, there is no guarantee that it will stand the test of time.

Over time, new questions may emerge, or users may interact with your app differently. For example, users may ask shorter questions as they get used to working with the question answering service through the app.

There are two main things we can do to improve the model:

• Use active learning in the Language Studio: An easy approach that analyzes predictions for you. Whenever questions seem too similar, the service concludes you need to provide more information to disambiguate between questions more clearly. Suggestions are provided, and you can review them in the Language Studio in the Review suggestions tab.
• Log diagnostics in Azure: A more advanced...