Topic Model for Graph Mining.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Graph mining is crucial for analyzing diverse data types like documents and molecular structures.
• Existing topic models fail on graph data due to the 'bag-of-word' assumption, limiting latent topic discovery.
• Current graph learning methods struggle to fully leverage the rich information within graph structures.

Purpose of the Study:

• To propose an innovative graph topic model (GTM) for effective latent topic discovery in graph-structured data.
• To address the limitations of standard topic models in handling graph data.
• To enhance both unsupervised and supervised graph learning through improved topic modeling.

Main Methods:

• Developed a novel graph topic model (GTM) specifically designed for graph-structured data.
• Utilized Bernoulli distributions to model the relationships (edges) between nodes in a graph.
• Integrated edge information into the latent topic discovery process.

Main Results:

• The proposed GTM successfully incorporates edge information for latent topic discovery.
• Experimental results demonstrate that GTM enhances graph representation for learning tasks.
• GTM outperformed Latent Dirichlet Allocation (LDA) in graph classification tasks when using discovered topics.

Conclusions:

• The novel GTM effectively uncovers latent topics in graph data by considering edge information.
• This approach improves the accuracy of graph representation for both supervised and unsupervised learning.
• GTM offers a significant advancement over traditional topic models for graph analysis.