Instagram implemented an AI algorithm to delete offensive comments 06.07.2017
Instagram implemented an AI algorithm to delete offensive comments

At the end of June, Instagram implemented an AI algorithm to delete spam and offensive comments. Artificial intelligence identifies inappropriate phrases by analyzing the context. It deletes even those comments where neutral words imply rudeness and keeps messages where swear words are a part of expression.

The algorithm of text classification DeepText was developed by Facebook. The system uses machine learning and the concept of “word embedding” – the principle used by human brain recognizing the speech meaning. So, DeppText thinks like people and can be taught.

In the second half of 2016, Instagram management decided to implement a new technology of text recognition to delete spam. Before that, users’ comments were analyzed by a team of moderators. DeepText took 80% of their work. The system analyzes semantics of every message and checks the account of a user who posted the comment. If the message was written by a person the user isn’t following the system will check whether it is not a spam. Similar comments are also considered to be advertising.

Having understood that the system works effectively (the amount of spam in comments decreased substantially), representatives of the social network decided to use DeepText to moderate rude messages and comments violating communication rules. The AI algorithm evaluates the text and decides whether the comment is appropriate. If the mark is higher than a particular point, the message is automatically deleted. Moreover, the author can see it but users cannot.

DeepText is now working with English texts only but in the future new functions will be added. If required, the algorithm can be turned off in settings.

The social network management believes that the AI system will help Instagram to become the friendliest place on the Internet.

AI-industry news and cases for you