Study On Key Approaches For Visual Representation And Analysis Of Graph Data

Graph data can be widely found in real life,such as financial trading networks,academic cooperation networks,social networks,etc.The exploration and analysis of graph data is helpful for people to understand the correlation patterns between entities,analyze the behavior of entities,and identify entities with abnormal behavior.With the advent of the era of big data,the scale of graph data is growing,and the difficulty of analyzing graph data is increasing.Visualization of graph data can help users explore the information and phenomena contained in the data interactively.Therefore,the visualization and visual analytics of graph data has been a research hotspot in the field of visualization in recent years.The complexity of processing and analyzing large-scale dynamic graph data is mainly reflected in:1)graph data contains various substructures,to which analysis tasks are related,such as pattern recognition,anomaly detection,etc.;however,algorithms related to substructures,such as subgraph matching,subgraph isomorphism,etc.,often have high complexity and thus cannot be completed in a short time.In applications where low hysteresis is required,substructure-related visual analysis poses a challenge.2)Processing large-scale graph data not only faces efficiency problems,but also increases the difficulty for efficient visual analytics.3)Dynamic graph data implies complex cor-relations,including multivariate correlations and correlation evolutions across time and space.4)In dynamic graph data,there are often some nodes that are different behavior from the other nodes,such as abnormal transactions in financial networks.For the analysis of the anomaly correlation in dynamic graph data,it is necessary to integrate human-machine intelligence into visual analytics to achieve practical application.This paper takes network data as the main research object,and discusses some key methods in the process of network visualization and visual analytics methods for network data.The main contributions of this paper are summarized as follows:· Substructure oriented visual representation and visual analysis of graph data.In the data space,nodes are represented as high-dimensional vectors based on the topology of the network data by means of representation learning.By using high-dimensional vectors of nodes,this paper proposes a substructure query method when users explore large-scale net-works based on the network structure and user specified substructures.We compare the performance of our query methods based on different vector representations.An interactive visualization system is implemented to support the visual query of substructures and verify the effectiveness of representation learning and representation learning-based query method in large-scale network.· Multi-level visual representation and visual analysis for large-scale graph data.We discuss visual representations suitable for large-scale network analysis in the visual space.This paper discusses the visualization techniques for large-scale network from two different perspectives,including the visualization of single-entity and multi-entities ego networks and the visualization of communities in networks.We design corresponding visualizations for the two tasks.At the same time,this paper implement an interactive visualization system to support users to explore and analyze large-scale network data interactively.· Visual representation and visual analysis of dynamic graph for multivariate correla-tions.In some network data,correlations among entities are constantly changing.Such change is often caused by a number of implicit factors.This paper firstly design a visual ana-lytics system to support the analysis of dynamics of correlations from the macro perspective,i.e.,the overall evolution patterns of correlations,based on the theory of strong ties and weak ties in sociology.Then we design a visual analytics system to support the analysis of micro evolution patterns of correlations,i.e.,the patterns of subsets of the correlations,based on a data structure called time-correaltion-partition tree.We design a tree-based visualization with glyph designs to enable users to interactively generate the tree structure and analyze the micro patterns of correlations.· Visual representation and visual analysis of dynamic graph for abnormal behavior.There are often some nodes with abnormal behavior in network data.The abnormal behav-iors can be identified by obeseving the evolution patterns of nodes and structures.In real-life scenarios,it is often very important to identify and analyze the behavior of these anoma-lous nodes.Based on a rare category detection algorithm,this paper designs a tree cutting algorithm for dynamic networks to enable users to analyze multiple classes with potential anomalous behaviors in a dynamic network and implements a visual analytics system with a novel visual design to support users identify and analyze anomalies in dyanmic networks.