I'm currently working on a multi-label multi-class text classification problem. In some part of the problem I need to give feedback to user that based on which features, words or phrases the algorithm decide to classify an instance to a certain category?
For example for a text like A B C D E F G H I J K L M N O P Q R I need output like this:
How can I consider this problem? what is the name or keywords of this problems? Is there any ready python package for this type of problem?
What you are looking for is interpretability in machine learning (ml) models. There is much discussion on this topic and is steadily gaining more attention, as more real-world applications in important subjects such as terrorism detection, precision-medicine etc. incorporate ml components.
There is at least one python module already doing what you seem to ask for, named lime.
If you are interested in the task in general, there are many other resources you can also check out (e.g. an extended presentation and article, conferences etc.).
