Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Uncovering the Interaction Equation: Quantifying the Effect of User Interactions on Social Media Homepage Recommendations

Hussam Habib, Ryan Stoldt, Raven Maragh-Lloyd, Brian Ekdale, Rishab Nithyanand

TL;DR

The paper investigates how user interactions shape homepage feed content across YouTube, Reddit, and X using sock-puppet crossover experiments. It introduces five feed composition vectors and a difference-in-differences framework to quantify treatment effects, revealing strong platform-specific personalization patterns: YouTube responds heavily to Like/View signals, Reddit to Join signals, and X shows selective responsiveness with political content deprioritized. The study also analyzes platform behaviors, including exploration-exploitation dynamics, reliance on explicit association signals, and dose-response effects, highlighting differences in network-driven versus algorithm-driven content propagation. Overall, the work contributes a methodologically rigorous audit framework and provides actionable insights into algorithmic personalization and transparency needs for modern social platforms.

Abstract

Social media platforms depend on algorithms to select, curate, and deliver content personalized for their users. These algorithms leverage users' past interactions and extensive content libraries to retrieve and rank content that personalizes experiences and boosts engagement. Among various modalities through which this algorithmically curated content may be delivered, the homepage feed is the most prominent. This paper presents a comprehensive study of how prior user interactions influence the content presented on users' homepage feeds across three major platforms: YouTube, Reddit, and X (formerly Twitter). We use a series of carefully designed experiments to gather data capable of uncovering the influence of specific user interactions on homepage content. This study provides insights into the behaviors of the content curation algorithms used by each platform, how they respond to user interactions, and also uncovers evidence of deprioritization of specific topics.

Uncovering the Interaction Equation: Quantifying the Effect of User Interactions on Social Media Homepage Recommendations

TL;DR

The paper investigates how user interactions shape homepage feed content across YouTube, Reddit, and X using sock-puppet crossover experiments. It introduces five feed composition vectors and a difference-in-differences framework to quantify treatment effects, revealing strong platform-specific personalization patterns: YouTube responds heavily to Like/View signals, Reddit to Join signals, and X shows selective responsiveness with political content deprioritized. The study also analyzes platform behaviors, including exploration-exploitation dynamics, reliance on explicit association signals, and dose-response effects, highlighting differences in network-driven versus algorithm-driven content propagation. Overall, the work contributes a methodologically rigorous audit framework and provides actionable insights into algorithmic personalization and transparency needs for modern social platforms.

Abstract

Social media platforms depend on algorithms to select, curate, and deliver content personalized for their users. These algorithms leverage users' past interactions and extensive content libraries to retrieve and rank content that personalizes experiences and boosts engagement. Among various modalities through which this algorithmically curated content may be delivered, the homepage feed is the most prominent. This paper presents a comprehensive study of how prior user interactions influence the content presented on users' homepage feeds across three major platforms: YouTube, Reddit, and X (formerly Twitter). We use a series of carefully designed experiments to gather data capable of uncovering the influence of specific user interactions on homepage content. This study provides insights into the behaviors of the content curation algorithms used by each platform, how they respond to user interactions, and also uncovers evidence of deprioritization of specific topics.
Paper Structure (23 sections, 13 equations, 2 figures, 8 tables, 3 algorithms)

This paper contains 23 sections, 13 equations, 2 figures, 8 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Platform affordances
  4. Content curation algorithms
  5. Homepage feeds and platform behaviors
  6. Experiment Design and Data Processing
  7. Experiment design
  8. Ethical considerations during data gathering
  9. Measuring changes in homepage feeds
  10. Measuring observed treatment effects
  11. Takeaways
  12. How do user interactions influence the topics shown in their homepage feeds?
  13. Methods for measuring the influence of treatment interactions
  14. Results
  15. How do user interactions influence platform behaviors?
  16. ...and 8 more sections

Figures (2)

  • Figure 1: Distribution of prominence of exploratory content on each platform in response to each treatment.
  • Figure 2: Cumulative effects of consecutive nearly-identical (topic, action) interactions on the TopicProminence homepage composition vector of each platform. \ref{['fig:dose:action:yt', 'fig:dose:action:X', 'fig:dose:action:Reddit']} show the average effects from each action, across all topics and \ref{['fig:dose:topic:yt', 'fig:dose:topic:X', 'fig:dose:topic:Reddit']} show the average effects from each topic, across all actions. A Sigmoid fit along with the mean square error is shown for each curve.