The social network said in an email to SocialTimes that some 60 million users visit On This Day every day, adding that 155 million users have subscribed to notifications from the feature.
On This Day allows users to look back at posts they shared and posts from friends that they were tagged in on the same day one year ago, two years ago, and so on.
However, while the feature brings back some fond memories for users, some not-so-fond memories have slipped through the cracks, and Facebook has taken steps to address that issue, including the introduction last October of two filters that allowed users to control the content they saw via On This Day by blocking specific users, dates or date ranges.
A Facebook spokesperson told SocialTimes last October:
We know that people share a range of meaningful moments on Facebook—from celebrating good times like a birthday to getting support in tough times like the passing of a friend or relative. As a result, everyone has various kinds of memories that can be surfaced—good, bad and everything in between. So for the millions of people who use On This Day, we’ve added these filters to give them more control over the memories they see.
On the anniversary of the official release of On This Day, a Facebook spokesperson offered more details in an email to SocialTimes:
We use a lot of different inputs to improve the On This Day experience. For example, we filter out content from exes and people you’ve blocked. These inputs would include things we know as a result of people using the breakup tools we launched back in November.
The social network also provided an overview of the artificial intelligence and machine learning work that goes on behind the scenes “to ensure that everyone using Facebook gets a meaningful, personalized experience”:
- Ranking: Facebook ranks and surfaces the most meaningful memories. These are ranked by an AI model developed at Facebook, trained in real-time (continuously learning); it gets better and more accurate at predicting what memories people want to see as they interact more with On This Day. Signals that Facebook uses to rank memories fall into: personalization (taking into account factors like previous interactions with On This Day, demographic info, friends they’ve blocked, etc.); and content understanding (essentially, the Applied Machine Learning Group-built Computer Vision and Machine Learning tech that uses annotated examples to train a deep convolutional neural network that recognizes visual concepts in photos—so Facebook can tell if a picture contains, say, a cat or a beach).
- User experience research: To better understand the types of memories Facebook should surface to people in News Feed, Facebook started by listening to people’s feedback directly through UX Research (via surveying people); Facebook then learned how to differentiate memories that are more meaningful than others, and how to tell when some memories that started off as pleasant became unpleasant.
- Filtering: Based on the UX Research, Facebook developed a set of filters for On This Day memories; these filters are tied to each person’s interactions and applied to memories surfaced in News Feed.
Readers: What have your experiences with Facebook’s On This Day feature been like?