Some research has shown that end-users will provide better labels, will have higher trust in the system, and communication will be improved if our system includes an explanation of how ORES arrived at its scores. We should provide these explanations whenever Jade is requesting judgment about a score.
The ideal explanation will include these elements (see the link above for details):
- Uncertainty level
- Sufficient low-level context
- ORES prediction
How to display the low-level context is an interesting problem and is the only non-trivial aspect of this task. We don't want to display all of the features, but only the ones most important to how ORES arrived at its score.
The key low-level features should be displayed in a way that the user can agree or disagree with each one. This can be split into future work if it turns out to be difficult.