Visual Analysis Of Multi-attribute Node Vector Network

Graph-based visual analysis has a wide range of applications in many important areas, such as social media, financial transactions, and biological networks. Effective graph layout algorithm is very important to visual analysis of network data in these areas. The traditional graph layout algorithm mainly considers the topology of a graph to generate visually-aesthetic layout results.With the rapid developments of Internet technology, the graph nodes usually contain multiple different data attributes, such as topics that people in social network are interested in, songs that people like to listen. We call the node of multiple attributes the node vector. The attributes of node vectors play an important role in revealing the relationships between nodes in network.With node attributes we can find meaningful node clusters and understand the role that nodes play in network. Most traditional graph layout algorithm does not consider multi-attribute node vectors, so the graph layout results can not reflect the influence of the node attributes. Aiming at this problem, in this thesis we propose the improved graph layout algorithm incorporating the embedded node attributes. Based on embedded node attributes graph layout, we further propose transfer function based framework for graph layout. The users can get the corresponding graph layout results by adjusting the transfer function. Specifically, users can set the parameters of the graph layout or select the node attributes through the transfer function. In order to calculate and display the results based on the transfer function interactively, we need to accelerate the calculation. In this paper we use Open CL framework for parallel computing to accelerate the graph layout. Finally, we implement a multi-attribute node vector network visualization system.The experimental results show that our system can not only help users detect the graph structures associate with the node attribute, but also can help us better understand the relationship between the role of nodes and the nodes layout.In summary, the main contributions of this thesis are listed as follows:1) Node attribute embedded graph layout algorithm: We design quadratic interpolation surface that maps the nodes’ attributes to parameters which control the force-directed layout. The layout result of this algorithm is associated with the value of the node attribute.2) Transfer function based graph layout algorithm: We extend the above node attribute embedded graph layout to transfer function based graph layout framework. Users can interactively change the graph layout result by adjusting transfer function. In this paper we adopt Open CL heterogeneous computing framework to accelerate the layout computation to enable the interactive operation of users.3) Multi-attribute node vector network visualization system: The input to the system is node vector network data. Through the computation of transfer-function based graph layout algorithm, the node-link layout graph is rendered on the main interface. The users can interactively adjust transfer function to obtain the desired graph layout. The system also provides zoom in/out, pan for users to explore the visualization. The experimental results of the real data, including Douban movie data and political blog data, verifies the effectiveness of our system.