Physics student uses machine learning to monitor social media sites
Anyone who has spent time online, from Twitter feeds to YouTube comments sections, knows that the internet is not to be confused with a realm of civil discourse. Hateful content can be prevalent, but Nikola Draca, a third-year physics student with a minor in computer science at the University of Ottawa, and his partner Angus MacLean, a student at McGill University, teamed up to address this growing problem.
Their solution comes in the form of an app called Soothe, which aims to block hateful content online by using machine learning techniques.
Draca said the concept goes back to his days in high school with MacLean. “We had a friend there that suffered from pretty severe anxiety, and she asked if we knew of a tool that did something similar to (Soothe)… At the time, and even now there’s only tools that let you block out very specific words, and it’s a really tedious process.”
But it wasn’t until recently that the idea really got off the ground.
“Last spring we were together at Carleton, at CUHacking, and we decided to just try and give it a shot and make this finally,” said Draca, explaining that he and MacLean wanted their project to do good for the community.
“Mental health awareness is a huge thing now, especially online harassment, so I think it’s an especially relevant tool right now,” he said. “And I think that the big platforms like Twitter and Facebook weren’t paying close enough attention to these issues until very recently.”
Draca says Soothe uses an algorithm to detect significant online harassment and block it from users’ browsers, using a process called sentiment analysis.
“This is an implementation of machine learning that just detects whether or not something has a positive or negative context,” he said. “We populated the original list of terms in each category ourselves, and then from that point on it searches for those terms in the context of the sentence it’s in, and if it’s overwhelmingly negative it gets rid of it.”
Draca said that Soothe now also contains a feature to flag newer slang that wasn’t initially taken into account, to make sure the app stays current. “Hopefully it’s going to remain relevant for a long time.”
“The response has been almost entirely positive,” he said. “Obviously I think it’s going to improve quite a bit as time goes on, but so far I think people have enjoyed using it.”
Draca said that in the future he wants Soothe to be able to work on images and video in addition to text, and that he’s hoping to grow Soothe to reach a wider audience.