Hate in the age of social media: using artificial intelligence to filter social media comments

Aggiornamento: 28 feb



  • In the previous month, 55,0 % of European users commented on a news story online, and 77,9 % read multiple comments at the same time.

  • According to GFK, Americans who leave comments on news stories, those who read them, and those who do neither, have a distinct socio-demographic profile. Whoever comments has a lower level of instruction and a lower level of income compared to those who read the comments.

  • Local newspaper websites and television stations, as well as social networking apps and pages, are popular destinations for both leaving comments and reading them.

  • 60.9% of those who comment or read the comments would want to see journalists respond to the comments, so advancing the cause of debate. Similarly, 58.7 % would like experts on the topic of the article to respond to the comments.


Despite these unequivocal findings from GFK research, the majority of newspapers ignore social media comments, which are frequently left unmoderated, allowing people with radically idea to use inappropriate, inaccurate, and dangerous content that may influence the opinions of other readers.


In recent years, Facebook has undoubtedly been the platform on which newspapers and Facebook itself have had the most difficulty managing reader comments and content. An examination of news comments and/or brand posts reveals that more than 20% of the comments concern:

  • Threatening and bullying comments

  • Use of profanity and swear words

  • Commercial and advertising scams

  • Spam that leads to malware threats

  • Fake information/fake news/ misleading articles

  • Unfiltered images that contain violence, pornography, content that promotes racism and hatred



While newspapers and brands in general can't avoid sharing their content on social media, the increase in comments that should be censored and the increase in the number of social platforms on which to share their content make the job of moderation arduous.


Types of social comment moderation


Pre-Moderation - This form of monitoring keeps unwanted comments from propagating on the internet. The user's comments will be assessed to see if it is acceptable and safe.

While it may take some time for posts to surface on social media networks, this filtering procedure assures that the material does not harm the newspaper's or brand's reputation.



Post moderation - The opposite of pre-moderation, it allows users to post or upload their content online in real time, and then filter out inappropriate content once it is discovered that those comments are inappropriate.



Reactive moderation - By clicking the "report button", dangerous, inappropriate content is seen by a moderator who can remove it. It is considered a more subjective form of comment moderation as more users may report content they disagree with even if that content is appropriate.

























Artificial intelligence to support comment moderation on social media


Analyzing hundreds if not thousands of comments per day on dozens of different social platforms is an impossible task without the support of tools dedicated to the cause.


One of the tasks that artificial intelligence is able to perform with amazing results is the classification of text across multiple categories. The AI algorithm is able to read thousands of comments per second, understanding their meaning and placing the text in the correct category (e.g. hate comment, vulgar comment, ok comment, etc.) or assigning each content a score that in the case of comments indicates the degree of "dangerousness" of the content.


Classifying comments into categories and/or assigning them a danger rating allows moderators to:

  • prioritize the most dangerous comments and approve all comments below a certain danger rating and/or in a certain category

  • automatically block all comments that exceed a certain danger level

How Roboticly.ai has enabled Socialbeat.io to automatically manage comments moderation


Socialbeat, in addition to automatically sharing content on social media, is able to track and aggregate, from multiple social media (Instagram, Facebook, Twitter, LinkedIn, Youtube, etc.) into a single dashboard all the comments that a newspaper and/or a brand receives on content posted by him.



These comments are then subjected to a "text classification" algorithm, developed with the Roboticly.ai no-code platform, that, based on the language and terms present, establishes the degree of danger of the comments and indicates on the dashboard those that require immediate attention.


The journalist or the social media manager can then intervene directly from the Socialbeat dashboard going to block the content, blocking the user from leaving future comments.




It's possible to outsource all content moderation to Socialbeat. In fact, Socialbeat, using Roboticly.ai employees, go to moderate content on behalf of the newspaper and / or brand, adding a layer of human classification to those comments that are more dangerous so as to go to remove them.


9 visualizzazioni0 commenti