Meta has published a new overview of how it works to improve your Reels recommendations, using user response surveys to better gauge which elements drive interest and engagement.
You’ve no doubt seen these yourself in the Reels feed, prompts that appear between videos and ask you what you thought of the Reel you just watched. Meta says it has adopted this approach widely and, based on the feedback provided, has gathered more information to help refine and improve Reels recommendations.
As explained by Meta:
“By weighting responses to correct for sampling and non-response bias, we have built a comprehensive data set that accurately reflects real user preferences. We go beyond implicit engagement signals and leverage direct, real-time user feedback.”
So instead of just using likes, shares and watch time as indicators of interest, Meta wants to go further and consider more elements that can further improve its recommendations.
And apparently it works.
According to Meta, before releasing these surveys, the recommendation systems only achieved 48.3% alignment with actual user interests. But now, after implementing the lessons based on these surveys, that has increased to more than 70%.
“By integrating survey-based measurement with machine learning, we create a more engaging and personalized experience by delivering content on Facebook Reels that is truly tailored to each user and encourages repeat visits. While survey-driven modeling has already improved our recommendations, there remain important opportunities for improvement, such as better serving users with limited engagement histories, reducing bias in survey taking and delivery, further personalizing recommendations for diverse user cohorts, and improving recommendation diversity.
This approach is not new; Pinterest, for example, describes how it is used similar studies to collect feedback to improve the recommendation systems.
But the rate of improvement is impressive, and it will be interesting to see if this leads to a significant improvement in the relevance of your Reels suggestions.
Although Meta is still behind TikTok in this regard.
TikTok’s mighty ‘For You’ feed algorithm remains the benchmark for compulsive engagement, allowing users to scroll through the app for hours.
So what does TikTok’s algorithm have that Meta doesn’t?
First and foremost, TikTok seems to have developed a better system for entity recognition within clips, giving TikTok’s system more data to match your preferences.
Yet TikTok is also very secretive about how its algorithm works, and won’t reveal much about this specific element, although we do know that TikTok’s system can identify very specific visual elements in clips.
Back in 2019, Intercepting it came across a set of guiding principles for TikTok moderators, including a set of very specific instructions for dealing with certain visual cues.
According to Intercepting it:
“[TikTok] has instructed moderators to suppress posts made by users deemed too ugly, bad, or disabled for the platform [as well as] videos with rural poverty, slums, beer bellies and crooked smiles. One document even goes so far as to instruct moderators to scan uploads for cracked walls and ‘notorious decorations’ in users’ homes.”
These guidelines were intended to maximize the ambitious nature of the platform, which would then drive more growth. TikTok admitted that such parameters once existed, but also clarified that these specific qualifications were never entered into TikTok itself, with the parameters copied from an earlier document intended only for Douyin, the Chinese version.
Although their existence indicates that TikTok can systematically detect these elements. I mean, you could assume that TikTok’s moderators wanted to manage this manually, and would reject videos with these elements based on human detection. But at the scale of the platform (both TikTok and Douyin have hundreds of millions of users) this would be an impossible task, rendering these notes completely useless. Unless the system can detect this via computer vision.
That’s where TikTok really wins, because it can understand a lot more about what you’re watching, and then factor that into your recommendations. So if you spend time watching a video of a blonde-haired, blue-eyed man, you can be sure you’ll see more content from similar creators.
Expand that to any number of physical features and background elements and you’ll see how TikTok can better suit your specific preferences.
So while TikTok also uses the more common matching, in terms of likes, watch time, etc., it also tries to keep users glued to their phones by adapting to their more primal preferences. And if the true depth of that process were ever made public, TikTok would likely come under intense scrutiny for using psychological biases and tendencies to coerce its users, possibly based on problematic and even harmful traits.
That’s where Meta is losing, because it can’t implement the same depth of knowledge to improve its systems. Theoretically, it could use more psychographic measures based on user history on Facebook, and among older users who have uploaded more of their personal data to the app, that could be effective. But mostly, Meta relies on more general algorithm signals, and now user surveys, to improve the Reels feed.
Are your recommendations looking better lately? This could be the reason, while it should also mean that your content is being shown to a more engaged audience.
#Metas #outlines #improved #role #recommendations