Computational Data and Social Networks

Thang N. Dinh, Minming LiLecture Notes in Computer Science 2023 · DOI · OpenAlex

This document represents the proceedings of the 11th International Conference on Computational Data and Social Networks (CSoNet 2022). Rather than a single study, it is a curated collection of research at the intersection of computer science, sociology, and optimization.

TL;DR

CSoNet 2022 explores the mechanics of "billion-scale" networks, offering new insights into how information, influence, and security operate in our hyper-connected world. From analyzing the "geography" of digital friendships to securing blockchain transactions, this volume showcases how computational tools are being used to solve complex social and technical challenges.

What Problem Does This Volume Address?

As digital networks grow to encompass billions of nodes, traditional analytical methods often break down. The researchers in this volume tackle the "scalability gap"—the challenge of understanding massive, dynamic systems where human behavior meets algorithmic intervention.

The core questions addressed include:

  • Information Integrity: How do we detect fake news and offensive speech across different languages and platforms?
  • Network Dynamics: How does physical reality (like geography) still constrain our "borderless" digital lives?
  • System Security: How can we protect decentralized systems like Ethereum or critical infrastructure like air transportation networks from targeted attacks?
  • Social Good: Can network science help us understand social stratification (like the caste system) or improve public health messaging?

Methods

Because this is a multidisciplinary collection, the methods are diverse, blending "hard" optimization with "soft" social science:

  • Graph Theory & Network Analysis: Using random walks for community detection and centrality measures to predict link formation.
  • Machine Learning & NLP: Leveraging sentence embeddings and Graph Neural Networks (GNNs) to predict "reposts" and categorize memes or offensive speech.
  • Optimization Algorithms: Applying Frank-Wolfe algorithms and submodular function maximization to solve "influence maximization" problems—essentially finding the best "seed" users to start a viral trend.
  • Simulation: Using Linear Threshold and Independent Cascade models to simulate how information flows across specific geographic regions.

Datasets

The papers utilize a mix of massive social datasets and niche technical benchmarks:

  • Social Media: A massive anonymized dataset from VK.com (the "Russian Facebook") with over 400 million users; Twitter datasets focused on COVID-19 polarization and vaping.
  • Specialized Data: Air transportation network logs, stock market indices during financial crises, and Ethereum blockchain transaction data.
  • New Benchmarks: The volume introduces specific datasets for Bangla fake news detection and anti-Asian offensive speech, addressing gaps in non-English NLP research.

Key Insights and Ideas

  • Geography Still Limits the Digital World: In a featured study on VK.com, researchers found that while we can connect with anyone, we mostly don't. The volume of friendship links drops sharply as physical distance increases, though this effect "saturates" after about 1,000 km.
  • The "Geographically Scale-Free" Property: Interestingly, the structural patterns of social networks (like degree distribution) look almost identical whether you look at an entire country or a single city. The "math" of friendship is fractal.
  • The Complexity of Polarization: Research into COVID-19 protests on Twitter suggests that polarization isn't just about "two sides"; it's driven by a complex web of local influencers and specific "trigger" events that shift network clusters.
  • Blockchain is Moving to "Layer 2": A significant portion of the security track focuses on zk-Rollups for Ethereum, suggesting that the future of blockchain research is less about the base layer and more about how to scale transactions securely.

Key Results

  • VK Study: Confirmed that geographic proximity effects on social media manifest primarily at short distances (under 1,000 km).
  • Optimization Gains: New heuristic algorithms for the "Student-Project Allocation Problem" and "Competitive Influence Maximization" show improved efficiency over standard greedy approaches in large-scale simulations.
  • Detection Accuracy: The volume reports high-performance benchmarks for detecting fake news in low-resource languages (Bangla) and identifying nuanced offensive speech patterns in social media conversations.

Limitations and Future Work

  • The "Snapshot" Problem: Many of the network analyses rely on static datasets (like the 2016 VK data). The authors acknowledge that social networks are highly temporal, and future work needs to focus on real-time, evolving graphs.
  • Privacy vs. Utility: As researchers move toward "billion-scale" computing, the ethical challenge of using personal social data—even when anonymized—remains a recurring concern that requires more robust differential privacy frameworks.
  • Cross-Platform Generalizability: Findings on one platform (like Twitter) often don't translate to others (like TikTok or VK). Future research aims to create "cross-platform" models of influence that aren't tied to a single UI or algorithm.

Summary updated

BibTeX
@inproceedings{dinh2023computational,
  author = {Thang N. Dinh and Minming Li},
  title = {{Computational Data and Social Networks}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2023},
  doi = {10.1007/978-3-031-26303-3},
}

Loading related content...

Figure