A big reason school IT Admins are drawn to Securly is our industry-first “Bullying and Self-Harm Detection”. This technology promises to keep students safe on social media by analyzing their posts and alerting school officials of incidents such as cyber-bullying, self-harm, and grief. In theory it sounds quite powerful, but how does our technology work, and more importantly, just how accurate is it?
We’ve come a long way since our successfully funded Kickstarter Project, in which we raised over $50,000 to aid in the research and development to perfect our technology. What makes our offering different from competing solutions can be summed up in a simple phrase: “sentiment analysis”. This means exactly what you are probably thinking. Instead of flagging on certain keywords in isolation, which tends to produce a high rate of false positives, our algorithms interpret the underlying emotion or sentiment behind the post and at that point decide whether or not it warrants being flagged. You can also read our original blog post to get a better understanding of the technical details.
Now on to the million dollar question – does it really work? If you were to ask us this same question a few months ago, we would have said yes, but with a small caveat – training our engine to become more sophisticated and reduce false positives is something that takes a substantial amount of time and effort. We’ve been fortunate enough to have two Securly team members working full-time on feeding our machine tens of thousands of social posts and rooting out false positives.
To best illustrate how far we’ve come, we’ll look at a before-and-after comparison. Below are some posts that we would have flagged in the early days of our technology. Notice that despite the excessive foul language, most of these posts are fairly harmless in their intent.
- “you the f***ing best”
- “eenie meenie miney mo you are nothing but a dirty h**. you think youre cute”
- “this s*** is stupid”
- “f*** you.”
- “i hate you so much”
- “sn sucks d****”
- “you f***ing hurt me
- “got 99 problems and a b**** aint one”
- “b**** keeps staring… better knock her shit off…”
- “i think you should shut the f**** up !”
- “@username pero theyre f***ing stupid clues”
Now let’s take a look at our flagged posts after the most recent update. You’ll see that the results are quite different now. Most, if not all, of these posts would warrant an alert to the IT Admin to take further action. If you’re looking at some of these posts and thinking, “wow, there’s no one word in there that would set off a red flag”, then you truly understand and appreciate the power of what our engineers have developed.
- “@username @username1 @username2 do y’all just pick on me bc im short? bc i can recall a good 2 times both of yall have tried to beat me up”
- “@username boy! i will punch you in the throat”
- “@username b**** i aint tryna say nothing. i didnt get farted on like some people…”
- “@username stfu [name]”
- “@username but you da h** im talking about”
- “@username u so mean to me”
- “you are a sl**”
- “go kill yourself”
- “since i have no friends to get high with.. guess i’m smoking alone tonight :)”
- “my mom has literally taken every f***ing thing that makes me happy and she wants me to like be hella cool to her. she rude as f***.”
- “hey guy sorry if i dont talk at all because my friend committed suicide”
- “have you ever felt like no one wants you in this world :(“
- “you’re really blessed if you have never had a physical addiction before. not to a person but to a substance. withdrawals will break you.”
- “do you ever get so heated that you feel like ripping off someones head and yanking their eyes out and stapling them on their knees? orrr..”
- “i’m ready to f***ing shoot myself”
- “i sit in the front of the class now and there is so much f***ing porn on my dashboard i hate my life”
- “my mom over here talking about god is going to punish me for cutting”
- “depression really sucks especially when you have no support. :(“
Why is this important? With so much reliance on device use in school and at home, students are increasingly turning to social media as an outlet to express themselves. If we think about almost all recent examples of extreme tragedy, the perpetrator has shown signs of troubling behavior, usually presenting itself through social media. And as we can imagine, there are potentially millions of student posts in schools across the globe that need attention but are otherwise going unnoticed. For us, we feel this is the aspect of the product that makes Securly truly secure. We were founded with the mission of protecting kids in all facets of their online life and continuing to improve our Bullying and Self-Harm Detection technology brings us closer to that goal.
Schools using Securly’s services have already noticed the powerful impact this technology can have on students’ well-being. Says Mark Nelson from Romeo Community Schools, “[We’ve] been very impressed with the language sentiment analysis (unavailable from other K-12 service providers); in fact, we used it just last week to inform a parent of an alarming post on Facebook by one of our students. Just one avoidance of a young person harming themselves or others would be worth a thousand times the subscription price.”
Our next roadmap items include: 1) Introducing this feature for users of iPads and our DNS service, and 2) Sending SMS texts to IT admins and parents with high-confidence alerts that warrant intervention. Have other ideas about what would make this offering even better? Please reply to this post and let us know.