Keeping online conversation constructive requires moderation, which takes time. We worked with The New York Times and Alphabet’s Jigsaw to build Moderator, a tool that uses machine learning to significantly increase the efficiency of human moderation.
Staff at The New York Times have manually moderated every comment posted to nytimes.com for years — yet they could only manage comments for approximately 10% of the site's articles. After implementing Moderator, efficiency gains have allowed journalists to open commenting on nearly every major article.
Under the hood
After closely assessing NYT’s needs, we built a new comment scoring tool using various machine learning models developed by Jigsaw and served through Perspective API. These models were trained on the Times’ previous moderation decisions, allowing Moderator to score and organize comments according to their predicted impact on conversation (e.g. likelihood of being inflammatory, off-topic, insubstantial, etc.). Staff could accept or reject Moderator’s findings, providing feedback that improves the API over time.
More, better dialog
By freeing the staff at The New York Times to quickly approve and publish comments on more articles, Moderator has removed a major bottleneck — time — allowing rich and constructive online discussion to flow unimpeded. And since the award-winning software is now open-source, it is expected to continue making the Internet a more constructive place for conversation across many other publishing platforms.
Read what The New York Times had to say about Moderator.