Protecting user experiences is at the core of every good product. No one wants to enjoy something once just to have a poor interaction the next time you go to use it. Here at Stack Overflow we are dedicating time and resources to the improvement of user experiences, particularly where malicious content and system distributions are involved. The newly established Moderation Tooling team for the Public Platform are implementing new tools and systems to keep users from the unwanted and interruptive exposure of spam and securing vulnerabilities from bad actors.
Like much of the internet, spam has long been an unfortunate reality on Stack Overflow. Spam has many looks but ultimately is unwanted and unsolicited advertisement or promotion that disruptions the Q&A experience. Moderation Tooling built an all-new spam filtering system to combat spam before it is ever published to Stack’s network. The idea behind the tool is simple: if a new post looks very similar to content that has been recently removed for being spam, it’s likely spam too.
In the past we’ve tried naive, or even legacy text comparison approaches by using a regex blocklist of words or phrases that often accompanied spammy posts. However, this approach was difficult to maintain, requiring engineers to manually notice trends and update lists accordingly. It was also incredibly brittle to strike the right balance between blocking a spammer from adding a phone number while still allowing a programming question about how to validate one.
The new wave of spam detection is here! By using vector embeddings and cosine similarity, the team has been able to build a tool that has an incredibly low false positive rate, or not mistakenly removing legitimate messages as spam. This has led to a 50% reduction in the time spam stays live on the platform and allotted valuable time back to our moderators to help maintain other parts of the platform’s integrity.
None of this would be possible without the dedicated members of our community helping us flag and identify what spam looks like on our platform. With special shouts to the folks behind Charcoal, who safeguard the positioning from unhealthy actors on a day by day, hourly, and even minute-by-minute foundation. We’re very excited to have the ability to leverage the onerous work they put in on figuring out spam and automate it earlier into the pipeline.
We’re taking over our efforts to enhance the positioning and enhance the instruments that our moderators have to keep up a wholesome community. In May, the Moderation Tooling team was formed to convey these moderator asks and requests to life, and in addition the options wanted to keep up a safer consumer expertise.
Our aim is to create a safe and constructive surroundings so that you can get your questions answered and share your information. With new techniques for stopping spam and detecting unhealthy actors, the Moderation Tooling workforce is dedicated to creating the platform safer, releasing up our group moderators and making certain higher Q&A for everybody. We consider {that a} clear, protected platform permits everybody to give attention to what issues most: studying and constructing. We invite you to get pleasure from a spam-free network experience whereas persevering with to ask, reply, and join with confidence on a more healthy, safer Stack Overflow.
