Short-Video Recommendations Need Content Depth, Not Just Engagement

Liwei Deng, Jing Jiang, Zhiwei Li, Yang Wang, Guodong Long· August 17, 2026 View original

Key takeaways

  • Current short-video recommenders prioritize immediate engagement, often at the expense of content depth.
  • The Content Depth Score (CDS) offers a new way to quantify cognitive stimulation in videos.
  • SCOPE-Bench is a new benchmark for evaluating content depth in recommendation systems.
  • Algorithms struggle to effectively recommend cognitively deep content, indicating a need for new approaches.

Who benefits

Social MediaEdTechMedia & EntertainmentDigital Advertising

Summary

A new study introduces the Content Depth Score (CDS) and SCOPE-Bench benchmark to evaluate short-video recommender systems beyond mere engagement, aiming to promote content that stimulates higher-order cognitive processes. Current systems consistently favor shallow content, and existing algorithms struggle to recommend cognitively deep content effectively.

Current short-video recommender systems primarily optimize for immediate user engagement, often leading to a preference for "shallow" content that quickly captures attention. This approach, driven by the attention economy, raises concerns about its long-term impact on users' cognitive engagement and mental well-being. To address this, researchers have proposed a new metric called the Content Depth Score (CDS), which quantifies how much a video stimulates higher-order cognitive processes using a seven-level scale based on cognitive psychology theories. They also introduced SCOPE-Bench, the first benchmark for evaluating content depth in short-video recommendations, featuring CDS annotations for 150,000 videos. Evaluations of 13 popular recommender systems using SCOPE-Bench revealed a consistent bias towards shallow content. Furthermore, algorithms designed to recommend cognitively deep content performed only marginally better than random selection, highlighting a significant limitation in current recommendation objectives and the need for new approaches.

Why it matters

Professionals in media, social platforms, and AI product development must consider the long-term impact of their recommendation algorithms on user well-being and cognitive engagement, moving beyond simple attention metrics.

How to implement this in your domain

  1. 1Integrate Content Depth Score (CDS) as a new metric alongside traditional engagement metrics in short-video recommendation system evaluations.
  2. 2Explore new model architectures or training objectives that explicitly optimize for content depth, potentially using SCOPE-Bench as a development tool.
  3. 3Conduct A/B tests to assess the impact of depth-aware recommendation strategies on user retention, satisfaction, and perceived value, not just immediate clicks.
  4. 4Collaborate with cognitive psychologists or educational experts to refine content depth definitions and annotation processes for specific platforms.

Original post by Liwei Deng, Jing Jiang, Zhiwei Li, Yang Wang, Guodong Long

"arXiv:2608.13990v1 Announce Type: new Abstract: Driven by the attention economy, short-video Recommender Systems (RSs) are primarily optimized to maximize user engagement by promoting videos that capture attention within seconds. These systems inherently favor shallow-content vid…"

View on X

Originally posted by Liwei Deng, Jing Jiang, Zhiwei Li, Yang Wang, Guodong Long on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses