[Paper Review] Social Network Integration: Towards Constructing the Social Graph

Yutao Zhang, Jie Tang|arXiv (Cornell University)|Nov 12, 2013

Complex Network Analysis Techniques25 references5 citations

TL;DR

This paper proposes a unified probabilistic framework for integrating multiple social networks into a single global social graph by linking accounts belonging to the same real person. It leverages cross-network consistency in social ties, user attributes, and logical constraints to achieve high-accuracy entity resolution, validated on LinkedIn, Videolectures, and ArnetMiner with strong integration performance.

ABSTRACT

In this work, we formulate the problem of social network integration. It takes multiple observed social networks as input and returns an integrated global social graph where each node corresponds to a real person. The key challenge for social network integration is to discover the correspondences or interlinks across different social networks. We engaged an in-depth analysis across three online social networks, AMiner, Linkedin, and Videolectures in order to address what reveals users' social identity, whether the social factors consistent across different social networks and how we can leverage these information to perform integration. We proposed a unified framework for the social network integration task. It crawls data from multiple social networks and further discovers accounts correspond to the same real person from the obtained networks. We use a probabilistic model to determine such correspondence, it incorporates features like the consistency of social status and social ties across different, as well as one-to-one mapping constraint and logical transitivity to jointly make the prediction. Empirical experiments verify the effectiveness of our method.

Motivation & Objective

• To address the challenge of integrating disjoint social networks into a cohesive global social graph.
• To discover accurate correspondences between user accounts across different social networks that represent the same real person.
• To develop a scalable, unified framework capable of handling arbitrary numbers of heterogeneous social networks.
• To evaluate the effectiveness of the method on real-world, large-scale social networks such as LinkedIn, Videolectures, and ArnetMiner.
• To explore the feasibility and benefits of constructing a complete, integrated view of individual social lives across platforms.

Proposed method

• The framework uses ad-hoc crawlers to collect data from multiple social networks, including user profiles and social ties.
• It formulates user identity linkage as a probabilistic inference problem, incorporating features like name similarity, profile consistency, and social tie overlap.
• The model enforces a one-to-one mapping constraint and logical transitivity to ensure consistent and valid linkages across networks.
• It employs a joint probabilistic model that combines social status consistency, tie pattern alignment, and topological features to predict cross-network correspondences.
• The integration results are fed back to adjust crawler priorities, enabling iterative refinement of data collection.
• The framework is designed to scale to large, heterogeneous networks and supports incremental integration.

Experimental results

Research questions

• RQ1 What social factors—such as user attributes or social ties—are most reliable for identifying the same person across different social networks?
• RQ2 To what extent are social network structures and user behaviors consistent across different platforms?
• RQ3 Can a unified probabilistic model effectively link user accounts across multiple social networks without unique identifiers?
• RQ4 How does incorporating one-to-one mapping and transitivity constraints improve the accuracy of user identity linkage?
• RQ5 What is the impact of feedback from integration results on the efficiency and coverage of data crawling?

Key findings

• The proposed framework achieves high accuracy in linking user accounts across LinkedIn, Videolectures, and ArnetMiner, demonstrating the feasibility of large-scale social network integration.
• Consistency in social ties and profile attributes across networks significantly improves the reliability of identity linkage predictions.
• The integration framework reduces false positives and maintains logical consistency through enforced one-to-one mapping and transitivity constraints.
• Feedback from integration results enhances crawler efficiency by prioritizing high-utility data sources.
• The method effectively handles the absence of unique identifiers and the heterogeneity of network topologies and semantics.
• Empirical results confirm that the framework can reconstruct a more complete and coherent global social graph from fragmented network data.