Auditing algorithmic responsiveness: behavioral signals, recommendation diversity, and platform difference on Chinese video platforms

Algorithmic recommendation reflects both system design and user behavior, but concerns about filter bubbles often focus on algorithmic curation. The extent to which recommendation diversity responds to the diversity of users’ behavioral signals, however, remains an open question. In this study, recommendation is treated as a trade-off between exploration (diversity) and exploitation (concentration) and a question is asked about how this trade-off shifts when user behavior is experimentally controlled. We conducted an agent-based testing audit on two Chinese video platforms, Douyin and Bilibili, collecting more than 400,000 videos from 35 agents per platform over 20 days. Agents browsing within a single interest area received less diverse recommendations than non-interacting agents on Bilibili and at the finer classification level on Douyin. Accounting for the predefined interest groups and for observations nested within agents, the association between input and output diversity holds within individual agents, though its magnitude is small. Overall recommendation diversity tracked input diversity on both platforms, though more strongly on Douyin; within a designated interest area, only Bilibili showed a consistently positive association between recommendation diversity and the breadth of an agent’s clicks. The personalized feed thus responds to the click-based interest signals supplied to it, in ways that differ across platforms and across levels of measurement.

Authors

Institutions

Publication Details

Journal
Chinese Journal of Communication
Published
2026-11-21
DOI
https://doi.org/10.1080/17544750.2026.2753764
Primary Topic
Recommender Systems and Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Auditing algorithmic responsiveness: behavioral signals, recommendation diversity, and platform difference on Chinese video platforms

Jinpeng Wang, Xin Yu, Mengye Yang, Peizhuang Miao et al.
Chinese Journal of Communication
Recommender Systems and Techniques
article

Auditing algorithmic responsiveness: behavioral signals, recommendation diversity, and platform difference on Chinese video platforms

Jinpeng Wang, Xin Yu, Mengye Yang, Peizhuang Miao, Yong Min
article en

Abstract

Algorithmic recommendation reflects both system design and user behavior, but concerns about filter bubbles often focus on algorithmic curation. The extent to which recommendation diversity responds to the diversity of users’ behavioral signals, however, remains an open question. In this study, recommendation is treated as a trade-off between exploration (diversity) and exploitation (concentration) and a question is asked about how this trade-off shifts when user behavior is experimentally controlled. We conducted an agent-based testing audit on two Chinese video platforms, Douyin and Bilibili, collecting more than 400,000 videos from 35 agents per platform over 20 days. Agents browsing within a single interest area received less diverse recommendations than non-interacting agents on Bilibili and at the finer classification level on Douyin. Accounting for the predefined interest groups and for observations nested within agents, the association between input and output diversity holds within individual agents, though its magnitude is small. Overall recommendation diversity tracked input diversity on both platforms, though more strongly on Douyin; within a designated interest area, only Bilibili showed a consistently positive association between recommendation diversity and the breadth of an agent’s clicks. The personalized feed thus responds to the click-based interest signals supplied to it, in ways that differ across platforms and across levels of measurement.

Chinese Journal of Communication
Shenzhen University (CN), Beijing Normal University (CN), Beijing Academy of Social Sciences (CN), Tsinghua University (CN)
Recommender Systems and Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon