Late Lucid Lectures Guild

Science, softly spoken.

Algorithms vs. Peers: Enhancing Engagement with Novel Content on Digital Platforms

Algorithms vs. Peers: Shaping Engagement with Novel Content

By Shan Huang, Yi Ji, Leyu Lin

DOI https://doi.org/10.48550/arXiv.2503.11561

Abstract

The pervasive rise of digital platforms has reshaped how individuals engage with information, with algorithms and peer influence playing pivotal roles in these processes. This study investigates the effects of algorithmic curation and peer influence through social cues (e.g., peer endorsements) on engagement with novel content. Through a randomized field experiment on WeChat involving over 2.1 million users, we find that while peer-sharing exposes users to more novel content, algorithmic curation elicits significantly higher engagement with novel content than peer-sharing, even when social cues are present. Despite users’ preference for redundant and less diverse content, both mechanisms mitigate this bias, with algorithms demonstrating a stronger positive effect than peer influence. These findings, though heterogeneous, are robust across demographic variations such as sex, age, and network size. Our results challenge concerns about “filter bubbles” and underscore the constructive role of algorithms in promoting engagement with non-redundant, diverse content.

Overview

This study explores how two main mechanisms—algorithmic curation and peer influence—affect how people engage with new or “novel” online content. “Novel” here is defined along two dimensions:

  • Non-redundancy: The content does not repeat what a user has seen before.
  • Diversity: The content covers topics that are different from a user’s historical interests.

Using a large-scale randomized experiment on WeChat (involving over 2.1 million users), the authors compare how these mechanisms influence both the type of content that is shown (exposure) and the type of content users actively click on (engagement).

Detailed Summary

1. Abstract

  • Purpose:
    The paper investigates the roles of algorithms (which tailor content based on individual behavior) and social cues from peers (like “likes” or endorsements) in exposing and encouraging users to interact with novel content.

  • Main Findings:

    • Exposure vs. Engagement: Peer-sharing tends to expose users to more diverse and non-redundant content (i.e., content that is new compared to what they normally see).

    • However, once the content is exposed, algorithmic curation leads to higher user engagement with that novel content, even when peer influence is also present.

    • Overall, while users naturally prefer familiar or redundant information, both algorithms and social cues help to counter this bias—with algorithms showing a stronger effect.

    • The findings are consistent across different groups of users (by sex, age, and network size).

  • Implications:
    These results challenge the common worry about “filter bubbles” (the idea that algorithms only show users what they already like) and instead point out that algorithms can play a positive role in introducing users to diverse content.

2. Introduction

  • Context and Motivation:

  • Digital Engagement:

    Modern digital platforms like WeChat use two primary ways to show content:

    1. Peer sharing: Where content is shared by friends or contacts along with social cues (e.g., a “like” or endorsement).
    2. Algorithmic curation: Where a computer program selects content based on a user’s past behavior.

  • Challenges Addressed:

    • Filter Bubbles: There is concern that algorithms, by focusing too much on past interests, might limit exposure to new topics.
    • Network Homophily: Peer-sharing might also limit exposure because friends tend to have similar interests.

  • Why It Matters:

    • Exposure to novel information is essential for sparking creativity, innovation, and informed decision-making in society.

    • Understanding these mechanisms helps in designing digital platforms that promote a broader range of information, potentially reducing information inequality and bridging social divides.

  • Research Gap:
    Previous studies looked at algorithmic and peer influences separately. This study is unique because it compares them within the same experiment on a single platform (WeChat).

  • Experimental Setup:

    • Groups:

      1. Group I: Received content that was selected by algorithms.
      2. Group II: Saw peer-shared content along with social cues (like visible “likes”).
      3. Group III (Baseline): Received peer-shared content without any social cues.

    • Measurement of Novelty:

      • Non-redundancy: Whether an article’s topic (represented by tags like education, society, etc.) overlaps with what the user previously engaged with.
      • Diversity: Measured using a concept called Shannon entropy (a statistical measure of unpredictability) to determine how spread out the topics are.

3. Key Findings (Including Insights from the Conclusion)

  • Content Exposure:

    • Peer-shared content (Groups II and III) tends to offer more novel content than algorithmically curated feeds (Group I). This is because friends naturally share a variety of topics based on their own diverse interests.

  • Content Engagement:

    • Despite the lower novelty in what is shown, users click more on novel content when it is delivered via algorithmic curation (Group I) rather than through peer-sharing. This indicates that even if algorithms show content that seems less diverse, they are more effective in encouraging engagement with new topics.

    • When social cues are added to peer-shared content (Group II), there is an increase in engagement—but still not as high as with algorithmic curation.

  • Users’ Natural Preferences:

    • The study confirms that users generally have a bias toward familiar or redundant content. This is explained by:

      • Selective Exposure: People tend to seek information that aligns with their existing beliefs.
      • Mere Exposure Effect: The more we see something, the more we like it, even if it is not new.

    • Both algorithms and social cues help counter this bias, but algorithms do so more effectively.

  • Demographic Variations:

    • The positive effect of algorithms was even stronger for certain groups (for example, males, middle-aged individuals, and those with larger social networks), while the influence of social cues was more pronounced among females, older users, and those with smaller networks.

  • Overall Conclusion:

    • Challenge to Filter Bubble Concerns: The findings suggest that rather than reinforcing narrow views, algorithms can actively encourage users to engage with a broader range of topics.

    • Practical Implication: Digital platforms might benefit from refining their algorithmic recommendations not only to tailor content to individual tastes but also to introduce diverse, novel content. This can help broaden user horizons without sacrificing engagement.

Explanation of Key Concepts

  • Algorithmic Curation:
    This is when a computer program selects and arranges content based on a user’s past behavior. For example, if you often read articles about technology, the system might prioritize similar content but can also include carefully chosen new topics that are slightly different.

  • Peer Influence and Social Cues:
    When content is shared by a friend or contact, and you see indicators like “likes” or comments, you might be more likely to check it out. This effect is rooted in the human tendency to trust recommendations from familiar people.

  • Novelty (Non-redundancy and Diversity):

    • Non-redundancy is about not repeating what you already know. For instance, if you always see news about sports, an article about science would be non-redundant.

    • Diversity means covering a range of different topics. A diverse set of articles might include pieces on science, art, history, and current events.

  • Shannon Entropy:
    Although this term comes from information theory, you can think of it simply as a way to measure how spread out or varied information is. Higher entropy means more variety.

Conclusion in Simple Terms

The paper concludes that while friends tend to share content that is already quite novel, the way algorithms are designed makes them even better at getting users to actually click on and engage with new types of content. This suggests that, rather than being a problem that creates narrow “bubbles” of information, well-designed algorithms can help broaden the scope of what we read and learn online.

The authors advocate that digital platforms should consider leveraging algorithmic curation as a constructive tool to encourage diverse engagement, ultimately supporting a more informed and innovative public.

More posts