The EU elections get underway tomorrow, and as debate heats up it’s likely the comments sections of major news sites will be ablaze with fiery opinions, harassment, and general internet nastiness.
To help counter abusive online comments in France, Alphabet think tank Jigsaw has launched French-language versions of Perspective and Tune, its AI-powered, troll-fighting technologies.
The Perspective API is first rolling out to the comments section of French news publisher Le Monde and will also be available to other publishers.
SEE: How to implement AI and machine learning (ZDNet special report) | Download the report as a PDF (TechRepublic)
The new version of Le Monde’s comment system going live today will use Jigsaw’s Perspective API to encourage readers to share views on news ahead of this week’s 2019 European Union parliamentary elections.
Perspective’s machine-learning model is designed to determine whether a comment could be seen as toxic to a discussion. The model was trained based on human responses to internet comments. Respondents rated the comments on a scale from “very toxic” to “very healthy”.
Perspective is already in use at the New York Times, the comment platform Disqus, and Spanish news publisher El Pais. Also available today is the French-language version of the Tune Chrome extension, which lets users customize how much toxicity they want to see in online comments.
Jigsaw is pitching Perspective as an alternative to online publishers shutting down their comments sections, as publishers often lack the resources to deal with the volume of comments.
Jigsaw says its Perspective AI gives publishers, writers and editors the tools to help them build a dialogue with their readers.
Using Perspective, The New York Times increased the number of articles with open comments by 300 percent, according to Jigsaw. Also, at El Pais, the toxicity of comments has declined by 7 percent on average. while overall commenting has increased by 19 percent.
El Pais uses Perspective to spot toxic posts and suggest changes to posters that would allow their message to pass its moderation system. It also forwards toxic comments to moderators, matching comments to the appropriate moderator depending on their level of experience.
Training the Perspective models to understand toxicity in French required analyzing “millions and millions” of comments, according to Jigsaw’s head of partnerships and business development Patricia Georgiou, and Jigsaw engineers Marie Pellat and Daniel Borkan.
The French comment data was sourced from Le Monde and other French-language news organizations.