Pixelfed Autospam has eliminated over 18K spam posts on pixelfed.social with an accuracy of 95%
The next step is to move beyond local and public only posts to non-public posts, remote posts and metadata like image alt text
We are sampling 5% of remote and non-public posts on pixelfed.social and logging detections (without taking action) for the next two weeks, the results will be made public later.
@pixelfed
5% feels like a pretty large false positive rate. What's being done to ensure the last 5% aren't needlessly labeled as spam?