Understanding Information Diffusion under Interactions

Yuan Su, Xi Zhang, Philip S. Yu, Wen Hua, Xiaofang Zhou, Binxing Fang

Introduction

In the modern social media landscape, information diffusion is not an isolated process. Every piece of content—or "contagion"—competes for a user's limited attention. Traditional models often treat these contagions as if they exist in a vacuum, or they rely on "black-box" latent factors that are difficult to interpret. This paper introduces the Interaction-Aware Diffusion (IAD) framework, designed to model the complex interplay between user social roles and explicit content categories (e.g., food, politics, or advertisements) to predict how information spreads more accurately and efficiently.

Methodology

The IAD framework decomposes the probability of information propagation into three distinct interaction types: User-Contagion (personal preference), User-User (social influence), and Contagion-Contagion (competition or reinforcement within an attention window).

The IAD framework architecture, detailing user role generation, topic extraction, and interaction inference.

To implement this at scale, the researchers utilized several key techniques:

  • Social Role Extraction: Users are categorized into three roles—Authority, Hub, and Ordinary—using network features like PageRank and HITS processed through a Gaussian Mixture Model (GMM).
  • Semi-Supervised Classification: Using a co-training approach, the model classifies contagions into 15 explicit categories (e.g., Economy, Food, Technology) by analyzing both content and user history.
  • Additive Probability Model: By using an additive formulation instead of complex multiplicative models, IAD significantly reduces computational overhead, making it suitable for massive datasets.

Data and Experiments

The study was validated using a real-world dataset from Weibo, a major Chinese microblogging platform. The dataset included over 19 million retweets across 140,400 popular blogs and involved more than 1 million distinct users. From this data, the team extracted 38 million "interaction scenarios" to train and test the model.

Results and Discussion

IAD demonstrated superior performance over state-of-the-art baselines like the IMM model in both F1-score and Accuracy. Most notably, IAD proved to be orders of magnitude faster, completing training in 6 hours compared to the 76 hours required by IMM on identical hardware.

Heatmaps illustrating inferred interactions: (a) User Role-Role, (b) User Role-Contagion Category, and (c) Contagion Category-Category.

The model's interpretability revealed several key insights into social media dynamics:

  • The "Food" Effect: Food-related content was found to be the strongest "attention-grabber," significantly suppressing the propagation of other contagions in a user's feed.
  • Status Gradients: Information flow is heavily influenced by seniority; Authority users are much more likely to adopt information from other Authority users than from Hubs or Ordinary users.
  • Zero-Sum Attention: The results confirm that social media is a competitive environment where most categories have a negative effect on each other, validating the theory of limited human attention.
  • Weakness of Ads: Advertisement-related contagions have the lowest capability to suppress other topics, suggesting a high level of user "immunity" to promotional content.

Conclusion

The IAD framework provides a highly interpretable and efficient method for predicting information diffusion by leveraging explicit social roles and content categories. While currently based on a sliding window of recent exposures, future work may explore how user roles evolve over time or vary across different sub-communities. The model's speed and clarity make it a valuable tool for both viral marketing and social science research.

Summary updated

Abstract

Information diffusion in online social networks has attracted substantial research effort. Although recent models begin to incorporate interactions among contagions, they still don't consider the comprehensive interactions involving users and contagions as a whole. Moreover, the interactions obtained in previous work are modeled as latent factors and thus are difficult to understand and interpret. In this paper, we investigate the contagion adoption behavior by incorporating various types of interactions into a coherent model, and propose a novel interaction-aware diffusion framework called IAD. IAD exploits the social network structures to distinguish user roles, and uses both structures and texts to categorize contagions. Experiments with large-scale Weibo dataset demonstrate that IAD outperforms the state-of-art baselines in terms of F1-score and accuracy, as well as the runtime for learning. In addition, the interactions obtained through learning reveal interesting findings, e.g., food-related contagions have the strongest capability to suppress other contagions' propagation, while advertisement-related contagions have the weakest capability.
BibTeX
@inproceedings{su2016understanding,
  author = {Yuan Su and Xi Zhang and Philip S. Yu and Wen Hua and Xiaofang Zhou and Binxing Fang},
  title = {{Understanding Information Diffusion under Interactions}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2016},
}
Figure