Information visualization of large data streaming

Abstract

The visualization of increasingly large streaming data has become a challenge for traditional static visualization algorithms and in the cognitive process of understanding features at various scales of resolution. In this thesis, we focus on stream visualization and study a class of visual analytic techniques that provide rich visual patterns to help common users better comprehend the main features of stream data. In stream visualization, we look for signifcant temporal patterns via several classical methodologies in the modern context of rapidly changing information content. However, the current methodologies for stream visualization are still limited. We summarize four critical problems for which the currently appropriate solutions require further improvements: (1) How to visually cluster useful patterns from streaming data; (2) How to smoothly map between data frames to provide a continuous visual effect in dynamic visualization and detect smooth trends and patterns; (3) How to automatically retarget the signifcant content of streams and make it suitable for different resolutions; and (4) How to visually query the patterns among streaming data To answer the frst question, we present two approaches to support the stream-based visualization and visual analysis. One approach is a density map estimation method to accurately cluster the high-density data. Another approach is a module-based clustering method to detect graph patterns and emphasize the interconnecting structures between local modules. Since the frst problem is fundamental, the answer to it provides the foundation for answering the other three questions. For the second question, we propose a novel algorithm called StreamMap that utilizes the diffusion model to smoothly morph frames among dynamic data. We also present a trend representation that can help convey the fow directions. The approach for the third question is a mesh-based energy optimization method. We propose a visual-saliency map to mark the regions with different signifcances. Auxiliary triangles are used to retarget the elements in visualizations, and the optimized result is achieved by solving a large sparse linear system. To answer the fourth question, we outline a new framework with two interactions for interactively querying the streaming data. We also designed a pair of representations to visualize and explore the query results. The effectiveness of the presented methods are demonstrated on several real datasets when dynamic visualization and a visual analysis of structured and unstructured patterns in streaming data are required.