PureRank: A Parameter-Free Recursive Importance Measure for Network Nodes

2024-12-31

AI Summary

PageRank relies on a manually tuned damping factor whose theoretical justification is weak and which fails to uniformly handle complex networks with signed (positive/negative) edge weights and bidirectional links. To address this, we propose PureRank—a parameter-free node importance measure grounded in a purely recursive definition that eliminates subjective hyperparameter dependence. Methodologically, PureRank introduces the first parameter-free recursive importance framework, ensures solution uniqueness and computational stability via strongly connected component decomposition, and integrates directed spectral graph theory with a signed-network splitting strategy to enable importance quantification on signed directed graphs for the first time. Experiments demonstrate that PureRank significantly improves parallel efficiency and large-scale scalability, while exhibiting strong robustness and broad applicability across diverse real-world networks.

Technology Category

• Machine Learning: Learning Preferences or Rankings
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Search and Optimization: Distributed Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

PageRank, widely used for network analysis in various fields, is a form of Katz centrality based on the recursive definition of importance (RDI). However, PageRank has a free parameter known as the damping factor, whose recommended value is 0.85, although its validity has been guaranteed theoretically or rationally. To solve this problem, we propose PureRank, a new parameter-free RDI-based importance measure. The term ``pure'' in PureRank denotes the purity of its parameter-free nature, which ensures the uniqueness of importance scores and eliminates subjective and empirical adjustments. PureRank also offers computational advantages over PageRank, such as improved parallelizability and scalability, due to the use of strongly connected component decomposition in its definition. Furthermore, we introduce the concept of splitting for networks with both positively and negatively weighted links and extend PureRank to such general networks, providing a more versatile tool for network analysis.

Problem

• PageRank
• Damping Factor
• Bidirectional Links

Innovation

• PureRank
• Strongly Connected Components Decomposition
• Network Bifurcation