Skip to content

Paper Bakshy2015 #
Find similar titles

Structured data

Author
URL

Polarization, Echo chamber, Filter bubble, Facebook

Table of Contents

Exposure to ideologically diverse news and opinion on Facebook #

... examined how 10.1 million U.S. Facebook users interact with socially shared news. We directly measured ideological Homophily in friend networks and examined the extent to which heterogeneous friends could potentially expose individuals to cross-cutting content. ...

Dataset and methods #

  • 10.1 million users with self-reported ideological affiliation

  • 7 million shared URLs between 20140707 and 20150107. They are categorized into "hard" (news and politics) and "soft" (entertainment, sports, ...) using Support vector machine on unigram, bigram, and trigram.

    • "bootstrapping" was used (different from the statistical method), meaning that they used regular expression to build a set of training labels.

  • 226,000 Hard URLs shared by at least 20 users with ideological affiliation.

  • content alignment \( A \) is the average ideological affiliation of each user who shared it.

Results #

  • there are many friendship links that cut across ideology. A liberal's fraction of conservative friends is ~0.2, the other way around is ~0.18.
  • For liberals (conservatives), 24% (35%) of hard news shared by their friends are cross cutting.
  • "potential from network" -> "exposed": slightly less (5% for conservatives and 8% for liberals) cross-cutting content after facebook's feed ranking.
  • "exposed" -> "selected": 17% risk ratio for conservatives and 6% for liberals.

Questions #

  • How may the choice of the user set (who self-reported their ideological affiliation) bias the results?
  • Fig. 1: How can the alignment score be large than 1 or smaller than -1? (the scale goes from -2 very liberal to 2 very conservative)

Incoming Links #

Related Articles #

Suggested Pages #

0.0.1_20140628_0