Curing a Toxic Comments Section with the Google and Jigsaw's PerspectiveAPI

A comment section is supposed to be useful. It’s supposed to help create a community that both you and your audience can learn from and enjoy. But we all know that’s rarely the case, what with the hordes of trolls and just plain unpleasant people corrupting your quest for meaningful dialogue.

Pablo Delgado, product director of Pisa Noticias, the publisher of the Madrid-based El País newspaper, recently described on Google’s The Keyword blog how the publication used the PerspectiveAPI to transform comment sections into a positive effort, rather than something to wrestle and wrangle. The API was created by Jigsaw and Google's Counter Abuse Technology team in a collaborative research project called Conversation- AI.

El País and Jigsaw worked to create a Spanish version of the free machine-learning tool. They analyzed those comments identified by its human moderators as toxic, which was a massive effort. The newspaper’s community interaction sections are extremely active, regularly receiving over 300,000 comments a month.

Now, the API now interacts with commentators when they leave a comment it identifies as toxic, using this new Spanish-language analysis. They receive a message suggesting they rethink and reword their comments into one that will pass moderation.

According to El País, the system is very successful, quieting the comments section and even growing usage by 19%. Drilling down further, the automatic moderation process that’s informed by this data analysis then forwards the more toxic comments to the most experienced moderators, while funneling the less toxic ones to those less-experienced mods.

El País now warehouses its comments, mining them to identify which commenters create toxicity as well as those authors or article topics associated with toxicity. The tool allows the newspaper to use this comments analysis to:

  • Run more articles likely to have a positive impact
  • Pinpoint when journalists should best engage in conversations with readers to create debate
  • Develop a commenter badges system and recognizing top comments
  • Create new products, such as a newsletter that highlights new (and reader-targeted) products

According to Delgado, the PerspectiveAPI restored the newspaper’s faith in the comments section and demonstrated its real value to El País and its readers. This initial Spanish version of PerspectiveAPI is available to anyone.

Original Article

How El País used AI to make their comments section less toxic