Tackling the spread of negative content on the Internet
According to a 2014 report on online harassment by Pew Research Center, 73% of adult internet users have witnessed someone being harassed online, while 40% have experienced it themselves. As is expected, the majority (66%) of those who have personally faced harassment have had their most recent incident occur on a social media platform. And even though all popular social media platforms provide tools for flagging negative content, only 8.8% of the victims have actually considered making use of these provisions. Essentially, passive techniques to curb the propagation of negative content like flagging have not proved to be very effective.
Combating negative content in real time
The aim of this project is to develop tools that can limit the creation of negative content. Instead of relying on passive techniques, we utilize artificial intelligence to actively intervene and restrain users from posting content that is hateful, abusive, racist, sexist or toxic in any other way. Currently, this is done with browser extensions that trigger warnings when users try to post content that is deemed negative by the artificial intelligence. Users can choose to react to these warnings or simply dismiss them. We record
the reactions in order to analyze the performance of the artificial intelligence.
Here is a demo (contents strictly for illustration purposes):
Backed by the state of the art deep neural network models
The artificial intelligence that powers this project comprises our state of the art methods in negative content detection. We employ natural language processing techniques based on deep neural networks to prepare models that can distinguish between negative and non-negative content with high accuracy. Since deliberate obfuscation of words and phrases by users has posed a significant challenge to automated detection in the past, we specifically designed our models to be resilient to this phenomenon. All the theoretical and implementation details of our methods will be published soon.
Always learning always improving
We provide for a simple mechanism by which you can report any negative content that you come across.
Each report that you make allows us to improve the artificial intelligence. This way,
we are able to keep up with the ever-evolving trends in internet jargon.
Here is another demo (contents strictly for illustration purposes):