Computational Data and Social Networks

Minh Hoàng Hà, Xingquan Zhu, My T. ThaiLecture Notes in Computer Science 2024 · DOI · OpenAlex

The 12th International Conference on Computational Data and Social Networks (CSoNet 2023), held in Hanoi, Vietnam, represents a pivotal moment in the evolution of network science. This volume of proceedings captures a significant shift from theoretical modeling toward high-impact applications in blockchain, misinformation detection, and domain-specific artificial intelligence.

Addressing the "Billion-Scale" Challenge

As social and computational networks grow in complexity, traditional analytical tools often struggle to keep pace. The research presented at CSoNet 2023 addresses the "billion-scale" challenge: the need to analyze, secure, and optimize massive, interconnected data systems. Key focus areas include:

  • Information Integrity: Developing robust methods for detecting deepfakes and tracking the spread of misinformation on platforms like X (formerly Twitter).
  • Domain-Specific AI: Creating NLP and recommendation tools specifically for the Vietnamese language and legal systems, where general-purpose models often fall short.
  • System Optimization: Solving complex combinatorial problems, such as the Traveling Salesman Problem (TSP) and vehicle-sharing logistics, under real-world constraints.
  • Infrastructure Security: Utilizing blockchain and IoT to secure sensitive supply chains, ranging from blood management to waste treatment.

Methodological Pillars

The proceedings are built upon four primary methodological foundations:

  1. Machine Learning & Prediction: Researchers utilized Graph Convolutional Networks (GCNs) for recommendation systems and compared Large Language Models (like GPT-3.5) against specialized local models (such as PhoBERT) for sentiment analysis.
  2. Optimization: The application of Mixed Integer Linear Programming (MILP), matheuristics, and submodular maximization proved critical for resource allocation and path interdiction.
  3. Network Analysis: Structural filtering and node entropy were employed to extract the "backbone" of complex networks, such as air transportation, and to identify anomalous actors.
  4. Blockchain & Security: The implementation of smart contracts and multi-chain interoperable systems (e.g., the MIDAS system) addressed pressing concerns in healthcare data privacy.
The editorial team for CSoNet 2023, representing institutions in Vietnam and the USA.

Specialized Datasets and Benchmarks

A hallmark of this year's conference was the introduction of specialized datasets designed to push the boundaries of regional AI:

  • ViEcomRec: A new dataset for recommendation systems in Vietnamese e-commerce.
  • VN-Legal-KG: A legal knowledge graph for identifying statutes within Vietnamese land law.
  • Twitter (X) Networks: Used for analyzing the behavior of misinformation spreaders.
  • African Swine Fever Data: A multiplex network approach to model disease spread in biological systems.

Key Insights: LLMs vs. Specialized Models

One of the most notable discussions in the proceedings involves the comparison between massive general-purpose models and smaller, linguistically-specialized ones. A study investigating whether GPT-3.5-turbo can outperform PhoBERT (a RoBERTa model pre-trained on Vietnamese) highlights the ongoing necessity for domain-specific architectures in low-resource language contexts.

Furthermore, research into "Secure Conversation" suggests that security in social networks is increasingly dependent on the physical layer of communication, not just encryption. The "cross-pollination" of social network analysis tools into biological sciences—exemplified by the modeling of African Swine Fever—demonstrates the versatility of these computational methods.

Results and Future Directions

The conference maintained high standards, accepting only 23 regular papers and 14 short papers from submissions across 20 countries. Reported results include quantitative improvements in optimization algorithms (such as more efficient A* Search for vehicle-sharing) and high accuracy in deep learning models for fake face recognition and social network stress measurement.

Looking forward, the proceedings identify the Internet of Things (IoT), blockchain modeling, and cloud computing as the primary frontiers for future research. While many high-performing models are currently optimized for Vietnam, their transferability to other low-resource languages remains a key area for exploration.

Summary updated

BibTeX
@inproceedings{ha2024computational,
  author = {Minh Hoàng Hà and Xingquan Zhu and My T. Thai},
  title = {{Computational Data and Social Networks}},
  booktitle = {Lecture Notes in Computer Science},
  year = {2024},
  doi = {10.1007/978-981-97-0669-3},
}

Loading related content...

Figure