Interest-Aware Content Discovery in Peer-to-Peer Social Networks

Lu Liu, Yonghong Guo, Lu Liu 0001, Yan Wu, James HardyACM Trans. Internet Techn 2018 · DOI

TL;DR

This paper introduces the Interest-Aware Social-like Peer-to-Peer (IASLP) model for efficient content discovery in decentralized online social networks (OSNs). By mimicking human social interaction strategies and building "interest" and "knowledge" networks, IASLP significantly improves content retrieval accuracy and reduces overhead compared to existing methods.

What Problem Does This Paper Address?

Online Social Networks (OSNs) have become ubiquitous, but their centralized architectures raise concerns about user privacy and system reliability due to single points of failure. Decentralized OSNs offer a promising alternative, but efficiently discovering relevant content within these large-scale, dynamic networks remains a significant challenge. Traditional content discovery methods, often designed for centralized systems or simpler P2P networks, struggle to cope with the complexity and scale of OSNs. This paper tackles the problem of finding useful information in decentralized OSNs by leveraging social network theories to create a more intelligent and adaptive content discovery system.

Methods

The core of the IASLP model is its "social knowledge index," which each peer node maintains. This index comprises two parts:

  • Interest Index: Stores knowledge about social contents from neighbors with similar interests, forming "content communities." This is based on strong relationships.
  • Knowledge Index: Stores knowledge directly relevant to search topics but potentially irrelevant to the peer's declared interests, forming "knowledge networks." This is based on weaker relationships.

IASLP employs several key strategies:

  • Self-Organization: Content communities and knowledge networks are formed autonomously by peer nodes based on their declared interests and search history, without explicit community building messages.
  • Interest-Aware Routing: When a peer receives a query, it first checks its local social content index. If not found, it uses its social knowledge index to identify relevant neighbors.
  • Adaptive Node Selection: Peers prioritize forwarding queries to neighbors with more matching social content, calculated through a "correlation degree" and "forwarding degree." This is inspired by concepts like "olfactory fatigue," where the perceived strength of a stimulus diminishes over time.
  • Probabilistic Forwarding: Queries are forwarded to neighbors within a content community with a high probability and occasionally to peers outside the community with a lower probability, balancing efficiency and breadth of search.
  • Knowledge Learning: Peers continuously update their social knowledge index based on search results, improving future routing decisions.

Datasets

Not discussed in the paper. The evaluation was conducted using a simulator.

Key Insights and Ideas

  • Social Theories as a Blueprint: The paper demonstrates that human social interaction patterns (e.g., forming communities based on shared interests, leveraging weak ties for new information) can be effectively mimicked to design more efficient P2P systems.
  • Dual-Network Approach: The distinction between "interest communities" (strong ties, shared interests) and "knowledge networks" (weaker ties, broader search) is crucial for comprehensive content discovery.
  • Adaptive Forwarding is Key: Dynamically adjusting the forwarding degree based on content relevance and correlation, rather than fixed rules, significantly improves search performance.
  • "Olfactory Fatigue" Analogy: The use of an adjustment factor inspired by olfactory fatigue provides a nuanced way to control how aggressively queries are forwarded, balancing recall and overhead.

Key Results

The IASLP model was compared against three relevant methods: RBFS, NeuroGrid, and SESD.

  • Higher Recall: IASLP achieved a 20% higher recall than SESD and significantly outperformed RBFS and NeuroGrid.
  • Lower Overhead: IASLP demonstrated a 10% lower overhead than SESD.
  • Performance in Dynamic Environments: The model showed good performance even with network churn (peers joining and leaving).
  • Effectiveness of Parameters: The study found that an adjustment factor (α) of 0.3 and a forwarding probability (p) of 0.9 offered optimal performance, balancing recall and overhead.

Limitations and Future Work

  • Scalability: While simulations were conducted, the authors suggest simulating the model in a larger network with High-Performance Computing (HPC) support.
  • Complexity of Human Interests: The paper acknowledges the diversification and complexity of human interests and proposes future work on analyzing these attributes to enhance semantic routing strategies.
  • Prototype Development: The authors plan to develop a prototype system in the Android environment for testing.
  • Privacy and Download Issues: The paper explicitly states that privacy issues and content download problems are not discussed.

Summary updated

BibTeX
@article{liu2018interest,
  author = {Lu Liu and Yonghong Guo and Lu Liu and Yan Wu and James Hardy},
  title = {{Interest-Aware Content Discovery in Peer-to-Peer Social Networks}},
  journal = {ACM Trans. Internet Techn},
  year = {2018},
  doi = {10.1145/3176247},
}
Figure