One may argue that it is one’s inalienable, human right to be stupid, but if you go online and call other people stupid, or insult them in a way that could be considered “trolling”, then Google’s new machine learning algorithm is likely to make a note of it.
The search giant has launched its Perspective API, which it claims analyses conversations and comments across a website and decides whether they are “toxic” or not.
One of the big problems for website publishers is if they allow comments from readers and visitors, they often get buried in spam. And if that wasn’t bad enough, there seems to be an army of people, or many armies of people, who seem to make it their business to go around posting offensive comments.
This often creates a hostile environment which may discourage reasonable debate and can quickly drag things down to the level where people are just hurling abuse at each other, although mostly it’s trolls attacking those who probably wouldn’t hurl insults back – they’ll just leave the website and not come back.
On its explanatory page, Google provides a few examples of comments which it ranks on a scale of “toxicity”, for which it uses a “toxicity model” developed by Survata – which basically involved asking people online to rank comments as “very toxic” or “very healthy” and a number of options inbetween.
Examples of toxic comments Google provides include:
- How can you be so stupid?
- They are liberal idiots who are uneducated.
- They have their heads up their ass.
These are probably the mildest “toxic” comments that you’d find on any large website. Usually the trolls use extremely strong language and not just in the general sense either – they’re often directed at other commenters and sometimes also target the writer or producer of the content below which the comments appear.
Early partners for Google’s Perspective API include:
- New York Times;
- The Economist; and
- The Guardian.
Google says: “Perspective is an API that makes it easier to host better conversations. The API uses machine learning models to score the perceived impact a comment might have on a conversation.
“Developers and publishers can use this score to give realtime feedback to commenters or help moderators do their job, or allow readers to more easily find relevant information.”
The search giant accepts that the API is not perfect. “It’s still early days and we will get a lot of things wrong.”
But as it improves it could become a useful tool for companies not only to create a more conducive environment for conversation and debate, it may also become a powerful tool for sentiment analysis.