Online communities play an instrumental role in bringing people together. At their best, online platforms are excellent places to enjoy, share, and connect. At their worst, they can become toxic, driving users away.
Which is why we’ve made it our mission to help Trust & Safety leaders maintain safe, thriving, healthy communities — and why we’ve built extraordinary technology: an astonishingly accurate system that identifies toxic behaviors like hate speech, radicalization, threats, and other ugly behaviors that drive users away from online communities.
But identifying disruptive behaviors is only half the equation.
The other half of the equation is still human judgment. Trust & Safety leaders need to be able to respond — at scale — to threats to their community while maintaining a close, personal connection to those communities; in short, remaining at the helm.
We’re excited today to introduce Guardian. Not everyone who cares about Trust & Safety thinks in code or datasets, so we built Guardian to be simple and non-technical, giving our customers the power to identify and respond to alarming behaviors any time, in real-time, and in context.
Simply put, Guardian is an easy user interface that sits on top of our behavior identification system. It gives Trust & Safety leaders the ability to digitize their guidelines, develop automated responses for certain behaviors and deliver more sensitive incidents to a team for review; and finally, to see how healthy their community is against industry benchmarks and their own KPIs.
Why Does It Matter?
Who cares if there is toxic behavior in an online community? Why bother enforcing guidelines?
Unbelievably, we get asked these questions a lot. Typically, by executives who prioritize growth over health (and don’t see how health leads to growth).
YouTube specifically lists Hateful Content as a disruptive behavior it doesn't support. So, were YouTube a customer, they’d purchase the following Behavior Identification models from us to support their Hateful Content policy.
Why the combination of models? Why not just Hate Speech? Because we’ve noticed that Hate Speech is closely connected to attempts at Radicalization (i.e. recruitment) and Threats. By combining these models, we get more accurate results.
Then, they’d use Guardian’s Automation Builder to detail responses taken when hateful content is identified. This nifty animation shows how Automations are built within Guardian: you can set complex conditions and add nuanced actions.
YouTube would, no doubt, automate some responses while directing some incidents to the Moderation Team for consideration. That team would work from Guardian’s Queue. The Queue displays entire incidents, in a prioritized list, so they can quickly assess context and make timely decisions.
Over time, YouTube’s Trust & Safety team would see how it’s responses (both automated and personal) affect the community with Guardian’s Analytics dashboard.
They may see, for example, that Hate Speech has been kept at required low levels, but they may also see trends in behaviors that would allow them to lower their own KPIs further. Perhaps the Analytics dashboard would show them that Hate Speech tends to increase during holy days for certain religions. Maybe they’d consider issuing preventive communications to their community in the days/weeks leading up to the days.
Wrapping It All Up
What gives Guardian an edge over other community health tools? Guardian is far more capable at detecting disruptive behavior than other technology on the market today. Learn more about how we detect behavior here.
It allows Trust & Safety leaders to automate what they feel comfortable automating the power to respond at scale while maintaining their connection to their communities.