Data stream and high-dimensional data visualization and the applications.

摘要

We can represent many real-world data sets as data streams and/or high-dimensional data. Major stock market such as New York Stock Exchange (NYSE) trades more than 8000 financial products daily, which generates massive trading time series in the form of data streams. NetFlow and DNS querying data generated by real-time computer network monitoring tools are usually large volume data streams or high-dimensional data sets. Computational manipulation of such data set is common. However, existing tools have been limited to their exploratory and analytical functionalities.; Visual depictions of graphs are external representations that exploit human visual processing to reduce the cognitive load of many tasks that require understanding of global or local structures. I argue that by using visual summarization and abstraction in display space and taking the characteristics of the user tasks into account, we can build visualization systems that are both effective and scalable.; This dissertation presents novel algorithms for visualizing data streams and high-dimensional data sets using image space compositing and supporting visual metaphors. To demonstrate the methodology for conducting visualization design and the utility of the proposed visualization algorithms and visual metaphors, I chose the application domain of computer network traffic and focused on leveraging human perception ability in the anomaly detection and analysis process. This thesis contains analysis of two network visualization systems for NetFlow and DNS Traffic data sets. Each of these two systems provides different novel views of the data, and we evaluate their efficacy with the intended tasks. I then generalize these findings in our analysis of the importance of summarization, interactivity, and specialization for the visualization systems to be effective and scalable.