Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media

Abstract: In a pre-registered algorithmic audit, we found that, relative to a reverse-chronological baseline, Twitter's engagement-based ranking algorithm amplifies emotionally charged, out-group hostile content that users say makes them feel worse about their political out-group. Furthermore, we find that users do \emph{not} prefer the political tweets selected by the algorithm, suggesting that the engagement-based algorithm underperforms in satisfying users' stated preferences. Finally, we explore the implications of an alternative approach that ranks content based on users' stated preferences and find a reduction in angry, partisan, and out-group hostile content, but also a potential reinforcement of pro-attitudinal content. The evidence underscores the necessity for a more nuanced approach to content ranking that balances engagement and users' stated preferences.

• The paper experimentally demonstrates that Twitter’s ranking algorithm significantly amplifies emotionally charged, divisive content compared to reverse-chronological timelines.
• The study finds that user satisfaction does not fully align with engagement-based content selection, particularly for politically sensitive tweets.
• The findings highlight the need for balanced ranking models that mitigate political polarization while fostering diverse perspectives.

Overview of the Study on Social Media Ranking Algorithms

The paper "Engagement, User Satisfaction, and the Amplification of Divisive Content on Social Media" investigates the effect of Twitter's engagement-based ranking algorithm on content selection and its implications for user satisfaction and sociopolitical discourse. The authors conducted a well-designed randomized experiment to quantify the impact of algorithmic content ranking against a reverse-chronological baseline. This study builds on prior, predominantly observational research, by providing robust experimental data.

The study involved 806 active Twitter users over two weeks, assessing their exposure to tweets ranked by the engagement-based algorithm versus a reverse-chronological order. Users rated these tweets based on a series of emotional and political attributes, allowing researchers to quantify the algorithm's influence on content emotionality, partisanship, and user satisfaction.

Major Findings

1. Amplification of Divisive Content: The engagement-based algorithm notably amplified emotionally charged and out-group hostile content. Political tweets ranked by the algorithm were more likely to express anger (62%) and out-group animosity (46%) compared to the reverse-chronological timeline (52% and 38%, respectively).
2. User Preferences and Satisfaction: Despite the algorithm's engagement optimization, it displayed lesser alignment with users' explicitly stated content preferences, especially regarding political tweets. Users exhibited a slight preference for the engagement-selected timeline overall but preferred chronological political content.
3. Sociopolitical Implications: The algorithm's preference for overtly emotional and partisan content potentially contributes to increased user polarization. Users showed worsened perceptions of their political out-group when exposed to the algorithm's content choices.
4. Alternative Ranking Approach Outcomes: The study explored an alternative content ranking method based on users' stated preferences. While this approach reduced hostility and anger in content, it raised concerns about reinforcing echo chambers by primarily reducing exposure to out-group content.

Theoretical and Practical Implications

The findings underscore the need for a balanced approach to social media content ranking. An algorithm solely focused on engagement metrics can inadvertently promote divisive content, potentially elevating political polarization and affecting societal discourse. Conversely, a model based strictly on users’ stated preferences might exclude cross-cutting perspectives, bolstering echo chambers.

From a theoretical standpoint, these results suggest a reconsideration of how engagement metrics are used to infer user preferences. The distinctions between revealed and stated preferences highlight the complexity in designing recommendation algorithms that thoroughly satisfy user desires without adverse social consequences.

Future Directions

Further research could extend the understanding of algorithmic long-term effects, particularly how they influence content production dynamics and contribute to feedback loops. Moreover, exploring methodologies that merge engagement data with direct user feedback could prove beneficial, providing a comprehensive understanding of user satisfaction metrics and their correlation with societal impacts.

In summary, this paper provides significant insights into the social and political ramifications of social media algorithms, presenting a compelling case for integrating user-centric and societal considerations into algorithm design. Its implications extend to both the development of future social media algorithms and the broader discourse around algorithmic influence on modern communication environments.