Fig 1.
Contingency tables of multidimensional annotations of political influencer Twitter/X accounts.
Influencer accounts are annotated based on the dimensions of Political Ideology (N = 886), Campaign Support (N = 356), Social Identity (N = 438), and Account Type (N = 2,129). Correlations among categories are examined for each dimension pair (a–f). A multivariate Fisher’s exact test is applied to each contingency table to assess dependencies between dimensions, with significance levels indicated. In Social Identity, Women denotes feminism or support for women’s rights; Religious, Black, and LGBTQ refer to self-identification or expressed support for the rights of these groups.
Fig 2.
Schematic diagram of network construction and community detection.
(a) The procedure used to construct the bipartite network between political consumer nodes (orange triangles) and political influencers nodes (blue circles). We project the bipartite network onto political influencers. The directed arrows between consumers and influencers represent the following relationships on Twitter/X. The undirected lines between influencers indicate their shared consumers, weighted by the number of shared consumers. (b) Illustration of multi-scale community detection in the projected influencer network. At the lower level, there are more communities, and the size of each community is smaller (shown in grey). At the higher level, the smaller communities merge, resulting in fewer, larger communities (shown in orange). (c) Histograms of community sizes at the five detected levels.
Fig 3.
Alluvial diagram showing the hierarchical organization of political influencer communities across five levels.
The different panels display the labeling of political influencer communities according to the following dimensions: Political Ideology (a), Campaign Support (b), Social Identity (c), and Account Type (d). The partition of the influencers in communities from a granular level to coarser levels is shown by the different grouping and arrangement of the bundles. We assign a color to a community if more than half of the influencer accounts within the community belong to one category. The transparency of the colors is proportional to the homogeneity of the labels; the higher the proportion of the majority category, the more salient the color. We only color communities when more than 30% of influencer accounts are annotated. In Social Identity, Women denotes feminism or support for women’s rights; Religious, Black, and LGBTQ refer to self-identification or expressed support for the rights of these groups.
Fig 4.
Projection of the consumer-influencer bipartite network onto both the consumer and influencer sides.
(a) Influencer communities projected on the consumer network. Each node represents a consumer who can follow single or multiple influencer communities. Consumers following the same communities are encircled. Consumer nodes that follow a single community are annotated in green, while those that follow more than one community are annotated in purple. All five levels are displayed. (b) Projected pie-chart graph of the influencer community network. The nodes are influencer communities, indicated by a pie chart of labels on four dimensions: Political Ideology, Campaign Support, Social Identity, and Account Type. Proportions of categories in each dimension are shown inside the pie chart. Edges between two nodes denote the consumers who follow both influencer communities. In Social Identity, Women denotes feminism or support for women’s rights; Religious, Black, and LGBTQ refer to self-identification or expressed support for the rights of these groups.
Fig 5.
Visualization of measurements of four community-level indices in the influencer community network.
(a) Identity Diversity, (b) Information Diversity, (c) Structural Integration, and (d) Connectivity Inequality. The nodes in each network represent detected influencer communities, and the edges represent the shared consumers between two influencer communities. The area of each node is proportional to the number of political influencers in the community. The color spectrum ranges from black to blue, with a darker color indicating more homogeneous ideological identities and domain information, more integration from other communities, and a more equal distribution of degree weights.
Fig 6.
Regression plot of significant individual attributes related to selective exposure patterns.
Regression analyses of consumers’ attributes on Demographics, News Consumption, Political Communication, Political Identification, Political Engagement, Perceptions of Incivility, Perceptions of Disinformation, Authority Trust, Populism, and Attitudes toward Democracy are conducted across five indices of selective exposure patterns: (a) Community Overlap, (b) Identity Diversity, (c) Information Diversity, (d) Structural Integration, and (e) Connectivity Inequality at four network community levels. The variables displayed are significant factors resulting from a forward selection based on BIC. The coefficient values are displayed in blue, and the confidence intervals at the 95% confidence level are shown with error bars. The red lines indicate the 0 value.