Labeled Subgraph Entropy Kernel

TL;DR

This paper introduces a novel labeled subgraph entropy kernel for graph similarity measurement, combining efficient enumeration and semantic enrichment to improve performance in real-world tasks.

Contribution

It proposes a new labeled subgraph entropy kernel with a dynamic programming algorithm, enhancing structural similarity assessment and computational efficiency.

Findings

Outperforms state-of-the-art graph kernels in experiments

Reduces time complexity through dynamic programming

Effectively captures semantic and topological information

Abstract

In recent years, kernel methods are widespread in tasks of similarity measuring. Specifically, graph kernels are widely used in fields of bioinformatics, chemistry and financial data analysis. However, existing methods, especially entropy based graph kernels are subject to large computational complexity and the negligence of node-level information. In this paper, we propose a novel labeled subgraph entropy graph kernel, which performs well in structural similarity assessment. We design a dynamic programming subgraph enumeration algorithm, which effectively reduces the time complexity. Specially, we propose labeled subgraph, which enriches substructure topology with semantic information. Analogizing the cluster expansion process of gas cluster in statistical mechanics, we re-derive the partition function and calculate the global graph entropy to characterize the network. In order to test our method, we apply several real-world datasets and assess the effects in different tasks. To capture more experiment details, we quantitatively and qualitatively analyze the contribution of different topology structures. Experimental results successfully demonstrate the effectiveness of our method which outperforms several state-of-the-art methods.