[Paper Review] Quantify and Maximise Global Viral Influence Through Local Network Information.

Yanqing Hu, Shenggong Ji+2 more|arXiv (Cornell University)|Sep 11, 2015

Complex Network Analysis Techniques3 citations

TL;DR

This paper proposes a local-information-based method to accurately estimate and maximize global viral influence in large social networks by leveraging percolation theory. It achieves brute-force accuracy with time complexity independent of network size, enabling efficient identification of optimal seed spreaders.

ABSTRACT

Measuring and maximising the influence of individuals is the key to understand and leverage the multiplicative potential of viral spreading over social networks. Yet these have been recognized as problems requiring global structural information, and extremely difficult for the enormous size and complexity of online social networks. Based on the percolation phenomenon, we show that they can be essentially reduced to local problems. Our estimated global influence of individuals based on local structural information matches that of brute-force search. Furthermore this measure leads to an efficient algorithm to identify the best seed spreaders for maximum viral influence. Its time complexity is independent of network size, making it practically useful for real networks. The underlying theoretical framework enables a thorough analytical assessment of the prediction accuracy and the absolute performance of the proposed methods.

Motivation & Objective

• To address the challenge of identifying influential spreaders in massive online social networks.
• To reduce the need for global structural information in influence maximization, which is computationally infeasible at scale.
• To develop a method that uses only local network information to estimate global influence with high accuracy.
• To create an efficient algorithm for seed selection that scales practically to real-world networks.

Proposed method

• Employs percolation theory to model information spreading dynamics across networks.
• Estimates global influence of nodes based solely on local structural features around each node.
• Derives a theoretical framework that links local network properties to global influence propagation.
• Uses a local approximation to simulate global influence without requiring full network topology.
• Designs an algorithm with time complexity independent of network size, enabling scalability.
• Validates the method through analytical assessment of prediction accuracy and performance bounds.

Experimental results

Research questions

• RQ1 Can global viral influence be accurately estimated using only local network information?
• RQ2 Is it possible to identify optimal seed spreaders without accessing the entire network structure?
• RQ3 How does the performance of the local method compare to brute-force global computation in terms of accuracy and efficiency?
• RQ4 What is the theoretical basis for the accuracy and scalability of the proposed local influence estimation?

Key findings

• The proposed local method achieves influence estimation accuracy comparable to brute-force global search.
• The algorithm's time complexity is independent of network size, enabling practical deployment on large-scale networks.
• The theoretical framework allows for rigorous analytical assessment of prediction accuracy and performance.
• The method enables efficient identification of optimal seed spreaders for maximum viral influence.
• The approach reduces the complexity of influence maximization from global to local, making it feasible for real-world applications.
• The results demonstrate that local structural information is sufficient to predict global influence with high fidelity.