Ranking by User Value: A Prosocial Design Recap with Smitha Milli
Engagement algorithms have negative effects. What are the alternatives?
Election cycles are timely reminders of what dominates discourse about social media: why does the most divisive, vitriolic content get uplifted most consistently on platforms? And perhaps more importantly, what can platforms do differently?
In other words, how can we (and the platforms we use) fix our feeds? This was the topic of conversation at the most recent Pro-Social with Smitha Milli, a postdoc at Cornell Tech who stopped by to share their research on using non-engagement signals in content ranking, and probe how we might move from optimizing engagement to ranking for user value in designing social platforms. The conversation was covered three core segments:
- Algorithms that rank by engagement force creators into producing more divisive content than they or users want.
When ranking content by engagement, the common indicators are likes, shares, or comments. Algorithms based on just those indicators, Smitha explained, tend to privilege the most hostile and divisive content, according to a study that gathered Tweets from a group of users and compared their engagement-based feeds and chronological-based feeds. Unsurprisingly, the engagement-based feed elevated content that was more polarizing; the users, in turn, walked away with a more negative perception of the political “other.”
More interestingly the users also reported that they did not actually want to see this type of content.
- Personalization is not the sole answer for moving away from engagement-based metrics on social media platforms.
Personalization, or adhering more intentionally towards users’ stated preferences, is one concept that Smitha explained has taken hold in design, academic, and technology circles. This approach attempts to find better signals than engagement to indicate stated preference; an excellent example being Meta’s “see less, see more” button (but one that has been relatively hidden until just recently). But other ideas aim to get at broader well-being outcomes, such as a survey asking “have you had a meaningful interaction on this platform?” or connecting long-term retention to proxy indicators; yet those strategies can be imprecise.
- Bridging-based ranking and value faithfulness
Is personalization the best approach for identifying what users find valuable, however? And should it be the only indicator in social communities? Smitha discussed the concept of bridging-based ranking, wherein platforms uplift content that has been identified as valuable by users with many different viewpoints. The question remains what type of content would receive this type of attention: would we see only puppies and hot air balloons, or truly substantive conversations?
Smitha ended with the emerging concept of value faithfulness, an inquiry into how we each indicate that we find something valuable. How does a click compare to a share? What about a “see less, see more?” This definition will continue to be explored in Smitha’s theoretical work.
Which of these approaches do you find most prosocial? Are there interventions you foresee taking off or facing challenges? Join the conversation on Slack! And check out the full conversation with Smitha below.
About the Prosocial Design Network
The Prosocial Design Network researches and promotes prosocial design: evidence-based design practices that bring out the best in human nature online. Learn more at prosocialdesign.org.