Frontiers in Physics (Sep 2024)

Heavy users fail to fall into filter bubbles: evidence from a Chinese online video platform

  • Chenbo Fu,
  • Chenbo Fu,
  • Qiushun Che,
  • Qiushun Che,
  • Zhanghao Li,
  • Zhanghao Li,
  • Fengyan Yuan,
  • Fengyan Yuan,
  • Yong Min,
  • Yong Min

DOI
https://doi.org/10.3389/fphy.2024.1423851
Journal volume & issue
Vol. 12

Abstract

Read online

Accelerated by technological advancements, while online platforms equipped with recommendation algorithms offer convenience to obtain information, it also brought algorithm bias, shaping the norms and behaviors of their users. The filter bubble, conceived as a negative consequence of algorithm bias, means the reduction of the diversity of users’ information consumption, garnering extensive attention. Previous research on filter bubbles typically used users’ self-reported or behavioral data independently. However, existing studies have disputed whether filter bubbles exist on the platform, possibly owing to variations in measurement methods. In our study, we took content category diversity to measure the filter bubbles and innovatively used a combination of participants’ self-reported and website behavioral data, examining filter bubbles on a single online video platform (Bilibili). We conducted a questionnaire survey among 337 college students and collected 3,22,324 browsing records with their informed authorization, constituting the dataset for research analysis. The existence of filter bubbles on Bilibli is found, such that diversity will decrease when viewing Game videos increases. Furthermore, we considered the factors that influence filter bubbles from the perspective of demographics and user behavior. In demographics, female and non-member users are more likely to be trapped in filter bubbles. In user behavior, results of feature importance analysis indicate that the diversity of information consumption of heavy users is higher than others, and both activity and fragmentation have an impact on the formation of filter bubbles, but in different directions. Finally, we discuss the reasons for these results and a theoretical explanation that the filter bubbles effect may be lower than we thought for both heavy and normal users on online platforms. Our conclusions provide valuable insights for understanding filter bubbles and platform management.

Keywords