User talk:Samwalton9 (WMF)

• User talk:Samwalton9 (WMF)/Flow

False positive reports of Automoderator

Hello, there is a discussion on zhwiki about the use of the false positive reporting page of Automoderator. The page is currently only for the community to acknowledge the issue and help users revert the false rollbacks made by Automoderator. Personally, I am thinking about whether it is possible to forward these reports to your team to help further improve the model, e.g. an on-wiki page that list confirmed false positives (like User:SCP-2000/sandbox/01).

Just my two cents and I am curious what do you think about this idea? Thanks. SCP-2000 (talk) 14:01, 14 February 2025 (UTC)

@SCP-2000 Thanks for the suggestion. Let me discuss this with the Research team, because my team can't really do anything with this data, but Research might be interested in using this data to improve the models. One idea we had was to automatically re-train the model based on false positive data (T337501), but we haven't been able to prioritise that idea yet. Samwalton9 (WMF) (talk) 14:04, 14 February 2025 (UTC)

Hello, I am wondering if there are any updates on this matter. Since the Multilingual model can be chosen for use soon, the community may want to report false positives to the Research team. Thank you! SCP-2000 (talk) 15:18, 15 November 2025 (UTC)

@SCP-2000 We didn't pursue this idea any further at the moment, but I know the Research and Machine Learning teams are in the process of figuring out how things like this could work in the future. I'll see if they have any updates and share them here if so. Samwalton9 (WMF) (talk) 13:47, 17 November 2025 (UTC)

@SCP-2000 I spoke with the Machine Learning team to learn more about this. At the moment they don't have the technical infrastructure or processes to retrain models based on the false positive data, but this is something they're interested in doing in the future. There are also other considerations with this, like how much to trust each false positive report, and what weight that should have in retraining the model. Those are things we could figure out, but it would take some time to make sure this is a reliable process. Samwalton9 (WMF) (talk) 11:02, 3 December 2025 (UTC)

Hello, thank you for your explanation and happy new year! There were around 18 false positive reports on zhwiki last year, and I think it's not difficult for the community to compile a list of false positives for the Machine Learning Team. I'm willing to help create such a list if needed. Please let me know if there's anything I can help with :) SCP-2000 (talk) 09:54, 11 January 2026 (UTC)