[Paper Review] Matching Community Structure Across Online Social Networks

Lin Li, William M. Campbell|arXiv (Cornell University)|Aug 3, 2016

Complex Network Analysis Techniques13 references6 citations

TL;DR

This paper proposes a multilayer community detection framework for aligning community structures across online social networks like Twitter and Instagram, using limited user identity mappings (seeds) as priors. It demonstrates that PageRank-based seed selection significantly improves detection accuracy, especially when shared user overlap is low, with oracle accuracy reaching 44.26% using only 8% of Twitter users as seeds.

ABSTRACT

The discovery of community structure in networks is a problem of considerable interest in recent years. In online social networks, often times, users are simultaneously involved in multiple social media sites, some of which share common social relationships. It is of great interest to uncover a shared community structure across these networks. However, in reality, users typically identify themselves with different usernames across social media sites. This creates a great difficulty in detecting the community structure. In this paper, we explore several approaches for community detection across online social networks with limited knowledge of username alignment across the networks. We refer to the known alignment of usernames as seeds. We investigate strategies for seed selection and its impact on networks with a different fraction of overlapping vertices. The goal is to study the interplay between network topologies and seed selection strategies, and to understand how it affects the detected community structure. We also propose several measures to assess the performance of community detection and use them to measure the quality of the detected communities in both Twitter-Twitter networks and Twitter-Instagram networks.

Motivation & Objective

• To detect a shared community structure across multiple online social networks despite unknown user identity alignment across platforms.
• To investigate how network topology—specifically the fraction of shared users—affects community detection performance.
• To evaluate and compare seed selection strategies (random, degree, PageRank) in improving detection accuracy with limited prior alignment.
• To develop and apply performance metrics such as variation of information, Jaccard similarity, and oracle accuracy to assess community detection quality.
• To explore the feasibility of using detected communities to enhance named entity disambiguation and large-scale graph matching.

Proposed method

• Uses three multilayer community detection methods: aggregation, linking, and relaxed random walk to combine network information.
• Applies an aggregation approach that normalizes adjacency matrices by weighted vertex degree to preserve structural fidelity.
• Employs seed-based community detection where known user mappings (seeds) guide the clustering process across networks.
• Uses PageRank centrality to prioritize seed selection, assuming high-centrality users are more informative for community alignment.
• Measures performance using variation of information (VI), Jaccard similarity matrix, and oracle accuracy based on known seed mappings.
• Performs experiments on both synthetic Twitter-Twitter and real-world Twitter-Instagram networks with varying seed fractions.

Experimental results

Research questions

• RQ1 How does the fraction of shared users between two social networks affect the accuracy of community detection across them?
• RQ2 How do different seed selection strategies (random, degree, PageRank) impact the quality of detected community structures?
• RQ3 Which multilayer community detection method (aggregation, linking, relaxed random walk) best preserves the underlying community structure?
• RQ4 To what extent can detected communities improve named entity disambiguation or large-scale graph matching tasks?
• RQ5 How does the quality and quantity of seed mappings influence the final detection accuracy, especially when ground truth is partially unknown?

Key findings

• The aggregation method consistently produced community assignments closest to the reference structure, outperforming linking and relaxed random walk.
• With only 8% of Twitter users as seeds (of which 44.26% were correctly assigned to the same community as their Instagram counterparts), PageRank-based seed selection achieved the highest oracle accuracy.
• PageRank-based seed selection outperformed both random and degree-based strategies, especially at higher seed percentages, with a 22.06% oracle accuracy at 7.3% seeds.
• Even with imperfect seed sets (due to basic matching methods), the PageRank strategy showed strong performance, suggesting its robustness to seed set noise.
• The results indicate that when shared user overlap is low, community detection becomes significantly harder, requiring more seeds for reliable results.
• The study confirms that high-quality seed sets are critical, and future improvements should focus on advanced entity resolution combined with network structure.