TGD: Topology-Enhanced Framework for Graph Drawing

Topological structure refers to the geometric properties of an object that remain invariant under continuous transformations. In the task of graph drawing, accurately capturing topological structures ensures that the layout of the graph aligns more closely with its inherent characteristics and relationships, thereby generating aesthetically pleasing layouts. Previous work has demonstrated that by optimizing models using graph topology-related loss functions, it is possible to learn sequences of operations that translate a graph into a layout. To further explore the model's ability to capture graph topological information, the enhanced topological representation framework for graph drawing, called TGD, is proposed in this paper. This framework employs Structural Deep Network Embedding (SDNE) with in-degree centrality encoding and spatial encoding during the graph preprocessing stage to enrich node embedding features, thereby enhancing the topological representation of the graph. By jointly optimizing multiple topology-related loss functions, the framework reduces the noise sensitivity of individual loss functions, allowing the model to better learn graph structure information and balance the overall layout optimization. Upon completion of the model training, it can quickly draw the input graph and achieve good visual representations. Experimental results show that TGD performs excellently in various graph layout tasks and has advantages over existing layout algorithms.