Personalization and Ideological Segregation in Social Media

Youzong Xu, Yunfei Cao, Ao Wang

Research output: Contribution to conferencePaperpeer-review


Social media often provides personalized content to enhance the efficiency and experience of their users. By modeling social media as a dynamic network, this study examines the effects of personalization on the opinion formation of social media users (henceforth, users). We argue that even when all users are openminded, personalization in social media can still lead to ideological segregation (e.g., political polarization). This ideological segregation will not disappear even if the population using this social media becomes very large. We characterize a necessary condition for the personalization algorithm used by social media not to cause ideological segregation. However, to satisfy this necessary condition in reality is almost impossible because that would require complete knowledge of users’ private information as well as an unchanging user population. To resolve the ideological segregation issue, we propose a random personalization algorithm that will not lead to ideological segregation. This algorithm can be easily applied in reality because it does not require any private information of the users or unchanging user populations.
Original languageEnglish
Publication statusAccepted/In press - Aug 2024
EventAsia Meeting of the Econometric Society, East & Southeast Asia 2024 - Ho Chi Minh City University of Banking, Ho Chi Minh City, Viet Nam
Duration: 2 Aug 20244 Aug 2024


ConferenceAsia Meeting of the Econometric Society, East & Southeast Asia 2024
Country/TerritoryViet Nam
CityHo Chi Minh City
Internet address


  • Ideological segregation
  • Social media
  • Personalization algorithms
  • Opinion dynamics
  • Dynamic networks

ASJC Scopus subject areas

  • Economics and Econometrics
  • Modelling and Simulation


Dive into the research topics of 'Personalization and Ideological Segregation in Social Media'. Together they form a unique fingerprint.

Cite this