Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Applying Machine Learning for characterizing social networks Agent-based models

Haoyuan Li, Lidia Conde Matos, Eduardo César Galobardes, Anna Sikora

TL;DR

Understanding the dynamics of large-scale social networks and enabling ABM-based simulations on HPC is addressed by a ML-driven pipeline that uses principal components ($PC_1$, $PC_2$, $PC_3$) for dimensionality reduction and K-means clustering to identify four user archetypes. The approach analyzes per-cluster attribute distributions with histograms and RSS-based goodness-of-fit tests, and it generates synthetic users to seed ABMs, validating the method on a Twitter dataset. The study identifies four archetypes—Influencers, Standard users, Sharers, and Lurkers—and provides a distribution-based generator that preserves cluster proportions and statistics, enabling scalable ABM on HPC. Overall, the work offers a practical, data-driven pipeline to synthesize realistic social-network populations for agent-based simulations and studies of information diffusion on HPC platforms.

Abstract

Nowadays, social media networks are increasingly significant to our lives, the imperative to study social media networks becomes more and more essential. With billions of users across platforms and constant updates, the complexity of modeling social networks is immense. Agent-based modeling (ABM) is widely employed to study social networks community, allowing us to define individual behaviors and simulate system-level evolution. It can be a powerful tool to test how the algorithms affect users behavior. To fully leverage agent-based models,superior data processing and storage capabilities are essential. High Performance Computing (HPC) presents an optimal solution, adept at managing complex computations and analysis, particularly for voluminous or iteration-intensive tasks. We utilize Machine Learning (ML) methods to analyze social media users due to their ability to efficiently process vast amounts of data and derive insights that aid in understanding user behaviors, preferences, and trends. Therefore, our proposal involves ML to characterize user attributes and to develop a general user model for ABM simulation of in social networks on HPC systems.

Applying Machine Learning for characterizing social networks Agent-based models

TL;DR

Understanding the dynamics of large-scale social networks and enabling ABM-based simulations on HPC is addressed by a ML-driven pipeline that uses principal components (, , ) for dimensionality reduction and K-means clustering to identify four user archetypes. The approach analyzes per-cluster attribute distributions with histograms and RSS-based goodness-of-fit tests, and it generates synthetic users to seed ABMs, validating the method on a Twitter dataset. The study identifies four archetypes—Influencers, Standard users, Sharers, and Lurkers—and provides a distribution-based generator that preserves cluster proportions and statistics, enabling scalable ABM on HPC. Overall, the work offers a practical, data-driven pipeline to synthesize realistic social-network populations for agent-based simulations and studies of information diffusion on HPC platforms.

Abstract

Nowadays, social media networks are increasingly significant to our lives, the imperative to study social media networks becomes more and more essential. With billions of users across platforms and constant updates, the complexity of modeling social networks is immense. Agent-based modeling (ABM) is widely employed to study social networks community, allowing us to define individual behaviors and simulate system-level evolution. It can be a powerful tool to test how the algorithms affect users behavior. To fully leverage agent-based models,superior data processing and storage capabilities are essential. High Performance Computing (HPC) presents an optimal solution, adept at managing complex computations and analysis, particularly for voluminous or iteration-intensive tasks. We utilize Machine Learning (ML) methods to analyze social media users due to their ability to efficiently process vast amounts of data and derive insights that aid in understanding user behaviors, preferences, and trends. Therefore, our proposal involves ML to characterize user attributes and to develop a general user model for ABM simulation of in social networks on HPC systems.
Paper Structure (11 sections, 3 figures, 2 tables)

This paper contains 11 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Research Methodology
  3. Data Preparation
  4. Clustering using Machine Learning
  5. Feature Analysis
  6. Preliminary Results: Use Case
  7. Step 1: Data Preparation
  8. Step 2: Clustering using Machine Learning Methods
  9. Step 3: Feature Analysis
  10. Conclusions
  11. Acknowledgments.

Figures (3)

  • Figure 1: General steps of the proposed methodology
  • Figure 2: Distribution of the numerical attributes in the different clusters
  • Figure 3: Probability plot of the original and generated users