Localized Algorithm of Community Detection on Large-Scale Decentralized Social Networks

Despite the overwhelming success of the existing Social Networking Services (SNS), their centralized ownership and control have led to serious concerns in user privacy, censorship vulnerability and operational robustness of these services. To overcome these limitations, Distributed Social Networks (DSN) have recently been proposed and implemented. Under these new DSN architectures, no single party possesses the full knowledge of the entire social network. While this approach solves the above problems, the lack of global knowledge for the DSN nodes makes it much more challenging to support some common but critical SNS services like friends discovery and community detection. In this paper, we tackle the problem of community detection for a given user under the constraint of limited local topology information as imposed by common DSN architectures. By considering the Personalized Page Rank (PPR) approach as an ink spilling process, we justify its applicability for decentralized community detection using limited local topology information.Our proposed PPR-based solution has a wide range of applications such as friends recommendation, targeted advertisement, automated social relationship labeling and sybil defense. Using data collected from a large-scale SNS in practice, we demonstrate our adapted version of PPR can significantly outperform the basic PR as well as two other commonly used heuristics. The inclusion of a few manually labeled friends in the Escape Vector (EV) can boost the performance considerably (64.97% relative improvement in terms of Area Under the ROC Curve (AUC)).

💡 Research Summary

The paper addresses a fundamental challenge in Distributed Social Networks (DSNs): how to detect a user’s community when each node possesses only limited local topology information, unlike centralized social networking services (SNSs) where the full graph is available. The authors propose a decentralized community detection algorithm based on Personalized PageRank (PPR), interpreting the PPR process as an “ink spilling” diffusion that starts from a seed node (the target user) and spreads through neighboring nodes while periodically “re‑starting” at the seed with probability α. This diffusion naturally captures the local affinity of nodes to the seed, allowing the algorithm to infer community boundaries using only the seed’s 1‑hop (and optionally 2‑hop) neighborhood.

A key innovation is the introduction of an Escape Vector (EV). In addition to the seed, a small set of manually labeled friends (typically 5–10) is added to the restart distribution. These labeled nodes act as trusted “ink sources,” biasing the diffusion toward the true community and away from peripheral or noisy regions. Empirical evaluation on a real‑world large‑scale SNS dataset (hundreds of millions of edges) shows that the EV‑augmented PPR achieves a 64.97 % relative improvement in Area Under the ROC Curve (AUC) compared with a baseline that uses only the seed. The method also outperforms two widely used heuristics—basic PageRank and label propagation—across precision, recall, and F1‑score.

From an implementation perspective, each node collects its immediate adjacency list and optionally the adjacency of its neighbors, builds a local transition matrix, and runs a few iterations of power iteration (or a fast approximation) to compute the PPR vector. The computational complexity is O(k·|E_local|), where k is the number of iterations and |E_local| is the number of edges in the local subgraph, making the algorithm scalable to millions of users without requiring global knowledge. Memory consumption stays within a few megabytes, enabling execution on mobile or lightweight clients.

The authors discuss several practical applications. In friend recommendation, the top‑ranked nodes in the PPR vector constitute high‑quality candidate friends drawn from the user’s own community. For targeted advertising, community‑aware profiling can improve relevance and conversion rates. Automated social relationship labeling can leverage the same diffusion scores to assign relationship strengths without manual input. Moreover, the method lends itself to Sybil defense: because Sybil nodes tend to form separate, loosely connected subgraphs, the PPR diffusion from a legitimate seed will assign them low scores, facilitating their detection.

The paper also outlines future research directions, including automatic construction of the Escape Vector from behavioral signals, handling dynamic graph updates for real‑time community tracking, and extending the approach to multi‑seed scenarios for detecting overlapping or hierarchical communities. Overall, the work demonstrates that high‑quality community detection is feasible in DSNs using only locally available information, and it provides a concrete, efficient algorithm ready for deployment in privacy‑preserving, decentralized social platforms.