Reproduce “Efficient Data Retrieval and Comparative Bias Analysis of Recommendation Algorithms for YouTube Shorts and Long-Form Videos” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Selimhan Dagtas, Mert Can Cakmak, Nitin Agarwal
Researchers developed an efficient, API-independent data collection framework to comparatively analyze recommendation algorithms for YouTube Shorts and long-form videos. They found that the Shorts algorithm exhibits a sharper increase in user engagement and joy, while both formats show an increasing trend in toxicity for outlier content, with multiprocessing achieving significant speedups up to 23x in data collection.
Key insights
- Recommendation algorithms, particularly for YouTube Shorts, exhibit distinct and often more immediate shifts towards high-engagement content compared to long-form videos.
- While algorithms tend to de-prioritize overtly angry content, they increasingly recommend content eliciting joy and surprise, potentially amplifying specific emotional states.
- To maximize engagement, both short-form and long-form algorithms may recommend outlier videos with increased toxicity, despite overall low mean toxicity, especially in user comments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.