Comparing information diffusion mechanisms by matching on cascade size

Area of Science:

• Information diffusion
• Network science
• Computational social science

Background:

• Information cascades are studied by analyzing structural properties like reach, depth, and speed.
• Comparing these properties across different content types is challenging due to interdependencies.
• Previous studies reported content-specific differences in cascade topology.

Purpose of the Study:

• To investigate the essentiality of controlling for cascade size when comparing structural differences between information cascades.
• To re-examine findings from notable studies on online diffusion of true/false news and various media types.
• To understand how cascade size influences observed structural properties in different diffusion models.

Main Methods:

• Revisiting datasets of Twitter cascades for true/false news and multimedia content (videos, pictures, news, petitions).
• Employing statistical methods that control for joint cascade statistics and cascade size.
• Analyzing classical diffusion models to identify conditions where structural differences reduce to size differences.

Main Results:

• For true- and false-news cascades, structural differences were largely explained by false-news cascades being larger.
• For videos, images, news, and petitions, structural differences persisted even after controlling for cascade size.
• Diffusion models show that structural property differences may or may not be reducible to size differences.

Conclusions:

• The underlying mechanisms of true- and false-news diffusion are likely similar, with differences primarily in spread rate.
• Cascade size is a critical factor to control for when comparing structural properties of information diffusion.
• Content type influences cascade structure, with some differences persisting independently of cascade size.