Tree-Based Kernel for Graphs With Continuous Attributes.

Area of Science:

• Graph Machine Learning
• Data Mining
• Computational Graph Theory

Background:

• Increasing availability of graph data with discrete or real-valued node attributes.
• Limitations of existing kernel methods for continuous attributes due to computational challenges.
• Need for efficient graph kernels that handle complex node features.

Purpose of the Study:

• To propose a novel graph kernel designed for complex and continuous node attributes.
• To maintain computational complexity similar to state-of-the-art methods while expanding the feature space.
• To develop an approximated variant for significant complexity reduction.

Main Methods:

• Feature extraction using tree structures derived from specific graph visits.
• Development of a novel graph kernel incorporating these tree-based features.
• Implementation and evaluation of an approximated version of the proposed kernel.

Main Results:

• The proposed graph kernel demonstrates superior performance on most of the six real-world datasets tested.
• The approximated kernel variant achieves classification accuracy comparable to state-of-the-art methods.
• The approximated kernel significantly reduces computational running times.

Conclusions:

• The novel graph kernel effectively handles complex and continuous node attributes.
• The approximated version provides a practical and efficient solution for graph data analysis.
• This research advances kernel methods for graph machine learning with continuous features.