A Ranking Approach on Large-Scale Graph With Multidimensional Heterogeneous Information.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Graph-based ranking is crucial for extracting value from structured data.
• Real-world graphs increasingly contain rich, heterogeneous information (node/edge features, prior knowledge).
• Existing methods struggle to efficiently utilize all available information, leading to issues like overfitting and high computational costs on large graphs.

Purpose of the Study:

• To address the challenge of large-scale graph-based ranking by effectively exploiting rich heterogeneous information.
• To develop a novel approach that improves ranking performance by integrating diverse data types within graphs.
• To overcome limitations of traditional link-based methods in handling complex graph structures and data.

Main Methods:

• Proposed a semi-supervised PageRank (SSP) approach.
• Integrated SSP within a unified semi-supervised learning framework (SSLF-GR).
• Simultaneously optimized model parameters and node ranking scores.

Main Results:

• The proposed method significantly outperforms existing algorithms on real-world large-scale graphs.
• Demonstrated the effectiveness of leveraging heterogeneous graph information for improved ranking.
• Achieved superior performance compared to methods that only partially consider graph information.

Conclusions:

• The developed semi-supervised learning framework (SSLF-GR) with SSP is an effective solution for large-scale graph-based ranking.
• Exploiting rich heterogeneous information is key to enhancing ranking performance.
• The method offers a significant advancement over traditional approaches in handling complex, large-scale graph data.