Community automated spam removal project updates
Hello. You may remember me from previous posts describing SmokeDetector and updates to the system to automatically apply flags to known spam. After the last post in March, there was a robust discussion regarding concerns, features and odds and ends that would make the community solution to spam on the Stack Exchange network even better.
Updates to the system since March 2018:
- SmokeDetector itself now provides a flag on all posts that are automatically flagged. This helps moderators to see that a post was flagged by the system and not only by community members.
- A dedicated RSS feed showing all posts that were automatically flagged and deleted, per site, is available. This can be accessed from metasmoke (see the blue RSS box underneth the graphs). This has been set up for moderation teams on several sites, so that it is pushed to a chat room for moderator review.
- The system has been casting up to 4 flags (SmokeDetector + up to 3 users) on posts that pass 99.9% historical confidence on spam reasons on a post.
- A user script (SIM - SmokeDetector Info for Moderators) has been written to expose autoflagging activity on a post. More details are available at the top of the previous post.
- metasmoke review system has been expanded to allow tagging domains found in spam posts. This has helped fight spam off of Stack Exchange as well, with several thousand posts removed across Wordpress, Medium, Weebly, Google Sites and others.
- System improved to ensure all automatically flagged posts are reviewed with feedback coming from multiple users, versus the minimium of only a single review previously.
- Improved the metasmoke dashboard and implemented per site dashboards to provide better visibility of the actions taken and results per site.
- Fixed a race condition that resulted in one post on Stack Overflow receiving 6 flags. On this post, the error was noticed in less than a minute and automatic flagging was stopped. Flagging remained off line for several hours while the issue was investigated and resolved. During that time, SmokeDetector remained online and reported potential spam via chatrooms as usual. The post being removed was spam and remained deleted, despite the error on our part in issuing too many flags.
Change in the near term:
We will increase the automatic flags cast on specific criteria. On those specific cases, we'll cast an additional flag (up to 5 total). These are a set of conditions that have 100.0% historical accuracy in determining whether a post is spam or not. This post is to provide transparency and solicit feedback. Charcoal intendeds to implement these changes on August 3, 2018.
Last paragraph: Just say what you're going to do, skip the Proposed
end with
Maybe set a date when those changes go in effect