Social Media Algorithms May Serve Content Users Dislike, Study Finds
Research published in the Proceedings of the National Academy of Sciences suggests social media algorithms may prioritize content that conflicts with users' values. The study indicates algorithms weigh replies more heavily, and users tend to comment on content they disagree with.
Social media platforms' algorithms may inadvertently expose users to content that clashes with their personal values, according to new research. A study published in the Proceedings of the National Academy of Sciences reveals that algorithms significantly weigh online posts that users reply to. The research indicates users are more likely to comment on content they take issue with than on content they agree with.
The study, which focused on the X platform (formerly Twitter), found that while algorithms presented content contradicting values to both Democratic and Republican users, this effect was more pronounced for Democrats. The analysis showed the algorithm favored posts related to upholding tradition, following rules, or keeping society safe, while demoting posts about looking after people, concern for those far away, being dependable, or protecting nature.
Researchers discovered that user interactions, particularly comments, serve as a stronger signal to the algorithm than simple likes. Because users tend to comment more on content that sparks disagreement, algorithms learn to serve more of this conflicting material. This creates a feedback loop that amplifies the presentation of value-clashing content in a user's feed.
Notably, the study found this tendency was stronger for Democrats compared to Republicans. The researchers suggest this is because Democrats object more strongly to content they reply to, thus creating a more potent feedback loop where the algorithm presents more clashing posts. The misalignment between content values and user values was over four times greater for Democrats than for Republicans.
The study's authors propose that platforms could improve alignment with user values by directly asking users about their values and tailoring feeds accordingly. They caution that algorithms optimized solely for engagement may increase polarization. Instead, surfacing bridging content that spans political lines while speaking to user values could foster both user autonomy and constructive conversation.