Simulating User Watch-Time to Investigate Bias in YouTube Shorts Recommendations

2025-07-06

AI Summary

This study investigates how transient user behaviors—such as swiping or skipping—in short-video platforms (e.g., YouTube Shorts) affect recommendation system properties, including topical continuity, semantic relevance, and content diversity. Method: We construct a large-scale dataset of 404,000 videos, simulate user interactions under multi-level dwell-time thresholds, and employ GPT-4o for fine-grained semantic relevance assessment along recommendation sequences. Contribution/Results: We propose a novel interdisciplinary semantic alignment framework integrating computer science and communication theory. Our empirical analysis reveals that dwell time critically modulates the trade-off between topical coherence and diversity; further, it induces systematic recommendation bias evolution—particularly pronounced in geopolitical topics—manifesting as content amplification, topical drift, and over-generalization. These findings provide cross-disciplinary, evidence-based insights to advance algorithmic transparency and inform platform-level content ecosystem governance.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Application Domains: Humanities & Computational Social Science
• Search and Optimization: Algorithm Configuration

Application Category

• User Modeling, Personalization and Recommendation: Studies of user behavior, including longitudinal effects of personalized systems
• Web Mining and Content Analysis: Content-based information diffusion
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Short-form video platforms such as YouTube Shorts increasingly shape how information is consumed, yet the effects of engagement-driven algorithms on content exposure remain poorly understood. This study investigates how different viewing behaviors, including fast scrolling or skipping, influence the relevance and topical continuity of recommended videos. Using a dataset of over 404,000 videos, we simulate viewer interactions across both broader geopolitical themes and more narrowly focused conflicts, including topics related to Russia, China, the Russia-Ukraine War, and the South China Sea dispute. We assess how relevance shifts across recommendation chains under varying watch-time conditions, using GPT-4o to evaluate semantic alignment between videos. Our analysis reveals patterns of amplification, drift, and topic generalization, with significant implications for content diversity and platform accountability. By bridging perspectives from computer science, media studies, and political communication, this work contributes a multidisciplinary understanding of how engagement cues influence algorithmic pathways in short-form content ecosystems.

Problem

• Investigates bias in YouTube Shorts recommendation algorithms
• Examines how viewing behaviors affect video relevance and continuity
• Assesses algorithmic impact on content diversity and platform accountability

Innovation

• Simulate viewer interactions with diverse topics
• Use GPT-4o to evaluate semantic alignment
• Analyze recommendation chains under varying watch-times