A Comprehensive Survey of Graph Embedding: Problems, Techniques, and Applications

Analysis

Why This Paper Matters

Graphs are ubiquitous in real-world data, from social networks to knowledge graphs, but traditional graph analytics often suffer from high computational and space costs. Graph embedding offers an efficient solution by mapping graph data into a low-dimensional space while preserving structural information and properties. This survey, published in IEEE TKDE in 2018, provides a timely and comprehensive overview of the rapidly growing field of graph embedding. With over 2000 citations, it has become a key reference for researchers and practitioners seeking to understand the landscape of graph representation learning.

The paper's significance lies in its systematic organization of the literature. By proposing two taxonomies—one based on problem settings and challenges, and another based on solution techniques—it helps readers navigate the diverse and often fragmented body of work. This structured approach not only clarifies the current state of the art but also highlights gaps and opportunities for future research. As graph embedding continues to be a foundational technique for knowledge graphs and beyond, this survey's impact is likely to endure.

Technical Contributions

• Formal definition: The paper provides a precise definition of graph embedding, establishing a common ground for subsequent research.
• Two taxonomies: It introduces a problem-centric taxonomy (categorizing challenges in different graph embedding settings) and a technique-centric taxonomy (categorizing solutions), offering a dual lens to understand the field.
• Comprehensive review: The survey covers a wide range of methods, from early matrix factorization approaches to contemporary deep learning-based techniques.
• Application summary: It consolidates applications such as node classification, node recommendation, and link prediction, demonstrating the practical value of graph embedding.
• Future directions: The authors identify four promising research avenues: computation efficiency, problem settings, techniques, and application scenarios, guiding future work.

Results

As a survey paper, it does not report new experimental results. Instead, its "results" are the synthesized insights from reviewing numerous methods. The paper successfully categorizes existing work, revealing trends and gaps. For instance, it notes that while many methods exist for static graphs, dynamic and heterogeneous graphs pose additional challenges. The proposed taxonomies enable comparative analysis and highlight that different problem settings often require tailored techniques. The identification of future directions has been influential, with subsequent research addressing some of the suggested areas, such as scalability and dynamic graphs.

Significance

The survey's significance extends beyond academia. By providing a clear framework and comprehensive overview, it lowers the entry barrier for newcomers and facilitates the adoption of graph embedding in industry. Its citation count attests to its role as a foundational reference. Moreover, the paper's emphasis on efficiency and application scenarios aligns with the practical needs of AI practitioners, especially in knowledge graph applications where embedding techniques are crucial for tasks like link prediction and entity resolution. The survey also helped shape the research agenda by outlining open problems, many of which have since been tackled by the graph neural network community. Overall, it remains a valuable resource for understanding the evolution and future of graph embedding.