Graph Representation Learning and Its Applications: A Survey

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Graphs are crucial for modeling real-world relational data.
• Graph representation learning maps entities to vectors, preserving structure and relationships.
• Numerous models have been developed over decades for graph representation learning.

Purpose of the Study:

• To provide a comprehensive overview of graph representation learning models.
• To cover traditional, state-of-the-art, and emerging techniques across different geometric spaces.
• To discuss practical applications, challenges, and future research directions.

Main Methods:

• Categorization of graph embedding models into five types: graph kernels, matrix factorization, shallow, deep-learning, and non-Euclidean models.
• Inclusion of graph transformer and Gaussian embedding models.
• Discussion of practical applications and domain-specific graph construction.

Main Results:

• A structured overview of diverse graph embedding models is presented.
• Key applications of graph embedding models are highlighted.
• Current challenges and future research avenues are detailed.

Conclusions:

• The paper offers a structured survey of graph embedding models.
• It emphasizes the breadth of techniques and applications in graph representation learning.
• It identifies open challenges and future research directions in the field.