Data-driven method of deciding number of contributions

[ppmedina11]

Hello everyone!

I was an audience before at Omran's presentation on fAIrSwipe in the State of the Map 2025 at Manila, Philippines. I remember Omran mentioning a problem with deciding how many contributors for a MapSwipe post is enough to constitute a push/submission. Correct me if I'm wrong, but I remember it being based on intuition or ad-hoc decision making in which in some MapSwipe posts, 5 contributors are enough, while in some, n contributors are needed.

On this end, I highly recommend creating a script which iteratively checks the reliability of the contributions in order to know whether the contributions are highly similar with each other, using existing statistical methods, and whether it is good to push these submissions due to them highly agreeing with each other. I think for me to help with this, I would need to know the following:

Problem Definition

• Can you further clarify the open question or problem? Is it more of a problem in terms of knowing how much contributions are needed or of evaluating the quality of the contributions?
• If it is the first (which is really my understanding), how do the maintainers/developers commonly decide in knowing whether the numbers of contributions are enough?

Understanding the Current Methods

• What are your criteria for accepting a contribution?
• Do we accept all contributions and do some specific geospatial methods to incorporate all of these contributions together? Or is the contribution checking very stringent or particular (i.e., we check each contribution's quality)?

Data Requests

• From my experience with MapSwipe, the types of contributions differ, ranging from tile validation to polygonal drawing. May I have all the types of contributions that are being done in MapSwipe?
• May I have some sample contribution data for each type, so that I have something to play around while aiming to provide a solution?

Thank you. Please tell me if you would be needing my email or any social channel, especially for the last request. For context, I am also working as a data scientist, so I do really enjoy working on problems which require statistical programming scripts which is relevant for this problem or even far more sophisticated deep learning scripts, which I don't think would be needed for this problem though.