We propose a method to achieve a balance between the Local and Global Structures (LGS) in graph embedding. This is achieved through the use of a tunable parameter. While some embedding techniques focus on capturing global structures, and others prioritize preserving local neighborhoods, only a few methods attempt to combine both aspects. However, it is challenging to effectively capture both local and global information in a two-dimensional graph representation, which is the common format for graph visualizations. The decision to use either a local or global embedding for visualization depends on the specific task and the underlying data structure, which may not be known in advance. In order to achieve a good balance between local and global structures in a given graph, the LGS approach aims to find the optimal solution. We evaluate the performance of LGS using synthetic and real-world datasets, and our results demonstrate that it is comparable to state-of-the-art methods. We use well-established quality metrics such as stress and neighborhood preservation, and introduce a new metric called cluster distance preservation to assess intermediate structure capture. All source code, datasets, experiments, and analysis are available online.
Finding the Right Balance: Local and Global Structures in Graph Embedding
