Branching process descriptions of information cascades on Twitter

TL;DR

This paper analyzes Twitter information cascades using branching process models, finding that a limited attention model better captures cascade dynamics than traditional independent cascade models, supported by analytical and empirical evidence.

Contribution

It introduces a branching process framework to model Twitter cascades and demonstrates the superiority of a limited attention model over the independent cascade model.

Findings

Limited attention model better reproduces cascade characteristics

Branching process models match empirical cascade statistics

Analytical results align with observed data

Abstract

A detailed analysis of Twitter-based information cascades is performed, and it is demonstrated that branching process hypotheses are approximately satisfied. Using a branching process framework, models of agent-to-agent transmission are compared to conclude that a limited attention model better reproduces the relevant characteristics of the data than the more common independent cascade model. Existing and new analytical results for branching processes are shown to match well to the important statistical characteristics of the empirical information cascades, thus demonstrating the power of branching process descriptions for understanding social information spreading.