Pinterest launches Today tab with curated topics, fights coronavirus misinformation with AI
In a bid to brighten the mood as the torrent of COVID-19 news continues unabated, Pinterest has announced the launch of Today, a new tab with curated topics and trending topic pins. While information from the World Health Organization and Centers for Disease Control about topics like handwashing is slated to feature prominently in the coming weeks, the Today tab will also show ideas for kid-friendly baking, work from home setups, family-favorite movies, comfort food recipes, and more.
Pinterest says the Today tab, which lives above the home feed within the Pinterest app for iOS and Android, will also spotlight content from guest editors and popular themes from world events, as well as topics people are searching for. Pinterest says this past weekend the platform saw more searches and pin saves than on any other weekend in its history and that searches for “homeschool schedule” and “freezer me” were up 32 times and 13 times, respectively.
Coinciding with the launch of the Today tab, Pinterest is bringing “compassion search” to the web, following the feature’s debut on Android and iOS last June. It’s a collection of emotional well-being activities created with the help of experts at Brainstorm and the Stanford Lab for Mental Health Innovation, with advice from Vibrant Emotional Health and the National Suicide Prevention Lifeline. Each practice offers ways to improve mood, from relaxation tools to self-compassion exercises.
Searching for things like “stress quotes,” “work anxiety,” or other terms indicating depression will surface compassion search activities, interactions Pinterest says are kept private and aren’t used to inform recommendations or ads. The company cautions that these suggestions aren’t meant to replace any form of professional care but says the activities might help users in need of support.
Lastly, Pinterest says it continues to use AI to fight misinformation about COVID-19 on its platform. To do so, it’s leveraging a machine learning system to identify pins it believes violate its health misinformation policy. A model trained using labels from Pinterest internally — which were supplied by the content safety operations team and Pinterest user-reported content — finds keywords or text associated with misinformation and blocks pins with that language while at the same time identifying visual representations associated with medical misinformation.
The model accounts for factors like image and URL and blocks any images online across Pinterest search, the home feed, and related pins, a Pinterest spokesperson told VentureBeat. “We have taken a firm stance against medical misinformation and have aligned those decisions with [our] model. For each image signature, the model produces a distribution of scores across the trained content safety categories, including porn, self-harm, hate speech, graphic violence, and medical misinformation, and a ‘safe’ class. Images with category scores above the corresponding threshold are filtered.”
Some of these same models helped reduce reported self-harm content on Pinterest by 88%, the company revealed in October 2019. They’ve also helped to remove such content 3 times faster than before.