For a later phase of ERI, create system for continual tuning of ORES scores.
Description
Description
Related Objects
Related Objects
Event Timeline
Comment Actions
Yes, that helps, thanks.
IIRC @Halfak said something about this data not being directly suitable for training (because it was influenced by ORES's decision, so there's a feedback loop), but he also said that storing false positives would be useful.
Comment Actions
Indeed. Storing false positives helps us notice trends and prioritize new work to make the models more effective.
Comment Actions
See https://www.wikidata.org/wiki/Wikidata:ORES/Report_mistakes for an example of how we usually work with misclassification reports.
Comment Actions
Essentially, this maps to Jade integration for ERI. See T209653: "Report error" button for ORES recent changes filter