Security and Privacy in Online Social Networks: A Survey

Sudarshan Kudlur Satyanarayana, Keshav Sood, Yuan Tao, Shui Yu

Abstract

Online Social Networks (OSNs) have evolved from simple communication platforms into massive repositories of sensitive personal data, making them prime targets for cyberattacks. This paper surveys the security and privacy landscape of OSNs, categorizing architectural vulnerabilities, human-centric social engineering threats, and the risks posed by third-party integrations. By analyzing centralized versus decentralized models and evaluating current defensive mechanisms, the survey provides a comprehensive taxonomy of the evolving threat vectors and the measures required to protect billions of users.

Introduction

The transition of OSNs from niche tools to global infrastructure has introduced significant security pitfalls. Features designed for ease of use—such as global search, seamless sharing, and third-party app ecosystems—are frequently exploited for financial gain and social harm. This paper addresses the lack of a unified taxonomy for these threats, focusing on architectural flaws, the exploitation of human trust, and the data-hungry nature of third-party applications.

Architectural Analysis and Methodology

The authors employ a systematic literature review to break down the OSN ecosystem into three key dimensions: architectural analysis, adversary modeling, and threat taxonomy. A primary focus is the comparison between traditional centralized architectures and emerging decentralized models.

Centralized OSN architecture where all user data is managed by a central server.

In centralized systems, as shown in Figure 1, a single provider manages all user data. While this allows for efficient global search and data availability, it creates a single point of failure for privacy. If the central server is compromised, the entire user base is at risk.

Decentralized OSN architecture utilizing storage nodes to distribute data control.

Conversely, decentralized or Peer-to-Peer (P2P) models (Figure 2) distribute data across various storage nodes. This approach offers users greater autonomy and reduces the risk of mass data breaches. However, the survey notes a "P2P Paradox": these systems struggle to replicate the seamless global search and high data availability that users have come to expect from centralized platforms like Facebook.

Threat Taxonomy and Insights

The paper organizes vulnerabilities into four distinct categories:

  • Privacy Threats: Such as the creation of "digital dossiers"—permanent, searchable records of a user's social history that can be used for blackmail or discriminatory practices.
  • Traditional Network Threats: Including DDoS and Cross-Site Scripting (XSS).
  • Identity Threats: Such as phishing and profile cloning. The survey highlights that phishing on OSNs is four times more effective than email-based phishing due to inherent social trust.
  • Information Retrieval Threats: Exploiting search functions to harvest data.

A particularly notable insight is the rise of Reverse Social Engineering (RSE). Unlike traditional "push" attacks, RSE involves building curiosity so that the victim initiates contact, effectively bypassing many automated security filters.

Defensive Mechanisms and Results

The survey evaluates several defensive techniques:

  • SybilGuard: Identified as an optimal protocol for defending against Sybil attacks (fake account creation) by leveraging the unique structure of social graphs.
  • XBook: A proposed flow-based model designed to restrict how untrusted third-party applications access and publish user data.
  • CAPTCHAs: Despite being a common user annoyance, they remain a critical recurring cost that deters automated profile cloning and large-scale RSE.

Conclusion and Future Outlook

While the paper provides a robust framework for understanding OSN security, it acknowledges significant challenges. Published in 2014, the scope does not include modern threats like AI-generated deepfakes or state-sponsored disinformation. Furthermore, while decentralized models offer privacy benefits, making them truly scalable and user-friendly remains a major engineering hurdle. Future research must focus on "smart clients" that can automatically enforce user-defined security rights, moving beyond the burden of complex, unread privacy policies.

Summary updated

BibTeX
@article{satyanarayana2014security,
  author = {Sudarshan Kudlur Satyanarayana and Keshav Sood and Yuan Tao and Shui Yu},
  title = {{Security and Privacy in Online Social Networks: A Survey}},
  journal = {EAI Endorsed Trans. Ind. Networks Intell. Syst.},
  year = {2014},
  doi = {10.4108/inis.1.1.e3},
}
Figure