Exciting update to our open source ranking algorithm that improves note quality, making it less likely that mediocre or unhelpful notes appear on Tweets. In the spirit of #AlgorithmicTransparency, here’s how and why it works…
Some notes appear on Tweets, then later disappear. As more ratings come in, they lose their Helpful status. This kind of self-correction is a good thing, but we consider these “false positives” — not necessarily bad notes, but ultimately not broadly helpful enough to show.
Analyzing data over the past few months, we’ve found that rating tags (for example, "sources not included") can be powerful signals to identify false positive notes early in the rating process. 
Going forward, notes that receive an atypically high number of such tags, relative to other Helpful notes, need to reach a higher level of helpfulness to show on Twitter.
The intuition behind this: if a note doesn’t cite a source, or might contain language that people see as biased or argumentative, it’s reasonable to hold it to a higher bar of quality before showing it.
To prevent this from being gamed, tags are inversely weighted by the statistical likelihood a rater would find the note helpful. If someone tries to game tags, their ratings will lose their intended effect.
In practice, we find that this significantly reduces the number of notes that appear on Tweets and then later disappear as they get noticed by more raters. The updates are live in GitHub at twitter.github.io/communitynotes…
• • •
Missing some Tweet in this thread? You can try to
force a refresh
