It has been reported that scientists have developed new machine learning algorithms which can successfully identify bullies and aggressors on Twitter with 90 per cent accuracy.
New algorithm spots Twitter bullies 'with 90% accuracy' https://t.co/QHStqJqo2h
— Sky News (@SkyNews) September 17, 2019
\”This is a very encouraging development that can help Twitter to advance their efforts in fighting cyber aggression and bullying.
What makes its application problematic though is that while 90% accuracy is great, it still leaves the 10%, or over 33 million of users, miscategorized. Innocent twitters are labeled as bullies, while the actual bullies are missed by the algorithm entirely, so at the end you cannot really do much with these results in an automated fashion and still have to manually review tens of millions of accounts.
Ideally, the algorithm should be pushed to minimize false positives to lessen impact on the innocent users. This way, even if the overall accuracy drops significantly since more bad actors will be missed by the algorithm, identified accounts of bullies and aggressors can be shut down automatically, which should help to remove significant number of offensive twitters. Absent that, the algorithm is still useful in prioritizing analysts\’ efforts to focus on the worst offenders, thus amplifying the efficiency of the account review process, and their feedback will gradually improve algorithm\’s performance and reliability.\”