Distinguishing mechanisms of social contagion from local network view (2406.18519v2)

Abstract: The adoption of individual behavioural patterns is largely determined by stimuli arriving from peers via social interactions or from external sources. Based on these influences, individuals are commonly assumed to follow simple or complex adoption rules, inducing social contagion processes. In reality, multiple adoption rules may coexist even within the same social contagion process, introducing additional complexity into the spreading phenomena. Our goal is to understand whether coexisting adoption mechanisms can be distinguished from a microscopic view, at the egocentric network level, without requiring global information about the underlying network, or the unfolding spreading process. We formulate this question as a classification problem, and study it through a Bayesian likelihood approach and with random forest classifiers in various synthetic and data-driven experiments. This study offers a novel perspective on the observations of propagation processes at the egocentric level and a better understanding of landmark contagion mechanisms from a local view.

• The paper demonstrates that Bayesian and random forest classifiers accurately distinguish contagion mechanisms using only local network data.
• It employs controlled experiments on star and Erdős-Rényi networks to compare performance when parameters are known versus inferred.
• The study validates its approach on empirical Twitter data, highlighting practical applications in marketing, public health, and misinformation control.

Distinguishing Mechanisms of Social Contagion from Local Network View

The paper, "Distinguishing mechanisms of social contagion from local network view" by Elsa Andres, Gergely Ódor, Iacopo Iacopini, and Márton Karsai, presents an analytical exploration into the classification of social contagion mechanisms based on egocentric network data. The researchers investigate scenarios where adoption mechanisms, typically categorized as simple, complex, or spontaneous, can induce contagion processes. These mechanisms drive the propagation of behaviors, ideas, or products within a network, and understanding their distinctions is crucial in contexts varying from marketing strategies to the diffusion of innovations or misinformation in social networks.

Introduction and Background

Previous works in the domain of social contagion have underscored the significance of both simple and complex contagion processes in describing adoption behaviors. Simple contagion models analogize the spread of behaviors to infectious diseases, whereby each peer interaction increases the probability of adoption. Complex contagion, on the other hand, posits that an individual’s likelihood of adopting a behavior depends non-linearly on the number of influenced peers, often crossing a threshold of social reinforcement. Additionally, spontaneous adoption mechanisms account for external stimuli independent of the existing social network, such as mass media influences.

One of the key challenges in distinguishing these mechanisms lies in the data availability. Most existing methods assume complete knowledge of the global network structure and the full unfolding spreading process, which is impractical in real-world scenarios where only local network information is typically accessible.

Methodology

The authors frame the differentiation of adoption mechanisms as a classification problem and explore this through a Bayesian likelihood approach and random forest classifiers across a series of experiments:

1. Experiment 1: This simple setup examines contagion processes on isolated star networks where neighbors undergo spontaneous adoption, while the central ego node follows either simple or complex contagion mechanisms. This highly controlled environment allows analytical approximation of classification accuracy.
2. Experiment 2: Increasing complexity, this experiment models contagion on larger Erdős-Rényi random networks, incorporating both neighbor-driven and spontaneous adoptions for each node. Researchers compare classification performance using both exact parameter knowledge and inferred values.
3. Experiment 3: Similar to Experiment 2, but without pre-assigned parameter knowledge. This experiment models a more realistic scenario where mechanism-specific parameters must be inferred, presenting a more challenging classification problem.
4. Experiment 4: This setup considers contagion on empirical Twitter networks, adjusting for temporal dynamics and incorporating activity-driven modeling. Nodes are assigned specific simple or complex contagion parameters derived from observed data, and spontaneous adoptions are integrated with event-time simulations to measure waiting times between exposure and adoption.

Results and Analysis

Across the experiments, several insights emerge:

• Experiment 1: Both Bayesian likelihood and random forest classifiers achieve high accuracy in distinguishing simple and complex contagion on isolated star networks. Exact theoretical estimation confirms the classification reliability, except when both contagion types exhibit rapid transmission (high β and low ϕ).
• Experiments 2 and 3: When expanding to larger and more varied network structures, both methods maintain significant performance, though a slight degradation in accuracy is observed when parameter values are not known a priori. The random forest classifier demonstrates robustness even with approximations, underlining the utility of machine learning in such classification tasks.
• Experiment 4: Introducing real-world complexities such as waiting times and temporal network structures, the random forest approach continues to yield meaningful classifications. This suggests practical applicability to empirical datasets like Twitter, although classification certainty reduces as network dynamism and individual heterogeneity come into play.

Implications and Future Directions

This work has several implications for network science and the paper of social contagion mechanisms. Practically, it highlights that even with limited, local network information, it is feasible to distinguish between competing contagion processes. The methodology can be adapted to various domains, such as targeted marketing, public health campaigns, or information diffusion control.

Theoretically, the findings underscore the nuanced roles of network structure and individual parameters in shaping contagion dynamics. Future developments could further refine these approaches by integrating more sophisticated temporal and multi-layer network models, as well as accounting for heterogeneities in external stimuli influences.

There is ample room for exploration beyond pairwise interactions, extending to higher-order interactions through simplicial complexes or hypergraphs. Investigating the propagation of different types of contagions, such as emotional or behavioral contagion, on multiplex networks could provide additional depth.

In conclusion, this paper provides a critical step towards understanding and differentiating contagion mechanisms using local network views. The integration of Bayesian and machine learning techniques showcases a robust analytical framework, opening avenues for better management and prediction of complex social contagion phenomena.