Research On Model-Driven Visual Analysis Of Network Data

Data visualization is an inter-discipline emerging in recent years,which involves data mining,computer graphics,human-computer interaction,artistic designing,visual analysis,etc.It aims at helping users to better understand,analyze and explore the data through translating the invisible and complicated data to a visible form like colors,shapes,textures and symbols using graphics and image technologies.With the coming of big data era,data visualization is a powerful tool to analyze data,which can quickly reveal the underlying behavior patterns and development trends of industries.However,big data is growing rapidly and often available in real-time,drawing from text,images,audio,video and so on,which makes it challenging to create an effective visualization method for big data and difficult to extract value from big data.Traditional visualization methods often have difficulty handling big data,including process and dynamic performance,model based data verification and user-preference driven hierarchical display.Most visualization methods are simple presentations of features like multidimensional,time-varying and huge amounts of data.There is no association and interaction among data model,analysis method and visual display,which makes it unable to explore the inherent law of big data.To address the problems above and satisfy the requirements of high performance,scalability and real-time analysis for big data,we propose a research method of model-driven visual analysis for network data that combines data rules,domain background,multi-scale display,and interactively embeds the models into the process of visual analysis to apply users' knowledge and experience to analyze network data and make decision to find more implicit rules and information.Mainly including:integrating the physical or mathematic model,knowledge rules and data analysis to build the data simulation model;designing effective visualization methods considering visual expression,interactive mechanism and other factors such as the user's perception to show the results generated by the data simulation model;interactive spiral exploring the data under the mutual cooperation and guidance of the front-end visualization method and the back-end simulation analysis model.Thus making visualization a tool to support process analysis,feature discovery and model optimization rather than a post-processing display.Therefore,this thesis introduces the research work according to the simulation analysis models of network data and their visual representations,which can be divided into the following several aspects.·Visual analysis of opinion propagation based on the physical model,that is applying the idea,concept,and method of physical model to simulate the social phenomenon of opinion propagation and reveal the rule of user behaviors with extending,correcting and improving these physical model.In this work,we propose a forum event propagation model based on the cellular automaton and a microblog retweeting model based on the dynamic fluid physics.The first model simulates the change of individual post number and individual viewpoint.The second model is to visualize the micro-blog retweeting by mimicking fluid dynamics.For them,we explore the internal mechanism and influence factors,and design the corresponding visualizations to enable users to analyze results,adjust parameters and verify models.·Visual analysis of collective effects based on the evolution model,mainly referring to exploring the organizing form,dynamic mechanism and path of cluster evolution by the application of visualization methods and interaction techniques.We propose a method to analyze the collective behavious based on the microblog retweeting visualization,and an emergency visualization based on the fire evacuation model.The first method applies the visualization to analyze the retweeting pattern and sentiment distribution among the users of different regions and different professions.Then explore the microblog retweeting characteristics of different events.The second method simulates the large-scale crowds in real-time and verify the evacuation data in complex environment through analyzing the characteristics of human behavior in emergent condition,considering the fire evacuation personnel in complex scene and the characteristics of evacuation behavior.·Interactive hierarchical visualization based on the geometric model,namely according to the topology of relational data,designing virtual energy system by fusing the attribute of nodes and edges into the layout,clustering and edge bundling of traditional network graphs.This methid applies interactions to achieve visual analysis from overview to details,further helps users quickly explore the relationships of data attributes and topology.In this work,a visual analysis method of adaptive hybrid scale is proposed for the multivariate graph.The hierarchical structure of the network graph is divided according to the structure and attribute information,and the roles of different attributes in the community structure are analyzed.Besides,a family tree model is applied for the multidimensional genealogy data.Different views are designed to analyze the demographic information,such as population migration,reproduction and family structure,to further explore the relations between family development,social background,geographical information and the natural environment,etc.Finally,our visualization methods are compared with traditional methods.We invite domain experts to evaluate and design the user experiment and questionnaires to collect feedbacks.A large number of experimental results demonstrate the usefulness of our methods in this paper.