Very interested in this for PersonalDashboard :)
T405105: Personalise the Recent Activity dashboard module
T404174: Surface edits to moderators which may require their review

The work was done. Check links on the task description.

Main Conclusions

• Predicting users' patrolling behavior (what revision their going to patroll next) is easier than predicting general editing behavior (what article they are going to edit next).
• There two main patrolling behaviors, depending in the nature of the patrolling task. Here we study two types of patrolling actions: Reverts and adding patrolling templates.
  • Content focused: People tends to revert more in articles they have edited previously. There just few users focused on reverts across multiple articles they haven't edited in the past.
  • Task focused: There users focused on adding templates related to certain policies. In this research we show there certain users that focus on adding/removing the {{peacock}} independently for the article.
• Therefore, our suggestion is to build two types of RecSyS, one task based and other content based.
• With these conclusions we can build a recommender system that finds editors that can benefit for getting personalized recommendations that combines their interests with outcomes of the ML-models for content integrity.

We got feedback on the datasets created in T408739:

Initial interest
Prior to any data being shared, interest in the project was high, with editors intrigued by the idea and stating that such a feature would likely be useful to them. Many had questions about how exactly the system would work. Questions included:

• Would it analyse both their content creation and patrolling activities?
• How far back in their editing history would it analyse?
• Would it be biased towards recent activity?
• Will it find edits to pages in topic areas they edit, as well as the specific pages they contributed to?

Testing feedback
Editors had feedback about the pages being selected for them to review. Although they thought it made sense to see edits to the precise pages they’ve edited in the past, many editors expressed a desire to see edits from other pages in the same topical area, or linked from the page they edited. Experienced editors are likely to already have pages they’ve edited on their watchlist, or otherwise be aware of the editing activity happening on them, so these recommendations were less useful.

• Can we incorporate pages that are connected to, or in the same topical area, as pages the user has edited?

Some pages were ones where the patroller had reverted vandalism on the page once, and they were otherwise not necessarily interested in the page in general.

• Can we select pages based on more significant editing experience, discounting or deprioritising pages where the user’s only edit is a single revert?

Response to the edits being presented was generally positive - most users said they thought the majority of edits were worth reviewing, even in cases where the edit itself didn’t need to be reverted.

A few specific highlighted issues that could be addressed in future iterations would be to not include:

• Edits made by experienced editors (e.g. other patrollers)
• Edits made by the user we’re generating suggestions for.
• Edits which have already been reverted.

An additional question for thought:

• Can we incorporate some of the parameters this model uses as configurable parameters so users can customize their patrolling experience?