Wendy E. Cowley

Visualizing sequential patterns for text mining

A sequential pattern in data mining is a finite series of elements such as A/spl rarr/B/spl rarr/C/spl rarr/D where A, B, C, and D are elements of the same domain. The mining of sequential patterns is designed to find patterns of discrete events that frequently happen in the same arrangement along a timeline. Like association and clustering, the mining of sequential patterns is among the most popular knowledge discovery techniques that apply statistical measures to extract useful information from large datasets. As out computers become more powerful, we are able to mine bigger datasets and obtain hundreds of thousands of sequential patterns in full detail. With this vast amount of data, we argue that neither data mining nor visualization by itself can manage the information and reflect the knowledge effectively. Subsequently, we apply visualization to augment data mining in a study of sequential patterns in large text corpora. The result shows that we can learn more and more quickly in an integrated visual data-mining environment.

Dynamic visualization of transient data streams

We introduce two dynamic visualization techniques using multidimensional scaling to analyze transient data streams such as newswires and remote sensing imagery. While the time-sensitive nature of these data streams requires immediate attention in many applications, the unpredictable and unbounded characteristics of this information can potentially overwhelm many scaling algorithms that require a full re-computation for every update. We present an adaptive visualization technique based on data stratification to ingest stream information adaptively when influx rate exceeds processing rate. We also describe an incremental visualization technique based on data fusion to project new information directly onto a visualization subspace spanned by the singular vectors of the previously processed neighboring data. The ultimate goal is to leverage the value of legacy and new information and minimize re-processing of the entire dataset in full resolution. We demonstrate these dynamic visualization results using a newswire corpus and a remote sensing imagery sequence.

Describing story evolution from dynamic information streams

Sources of streaming information, such as news syndicates, publish information continuously. Information portals and news aggregators list the latest information from around the world enabling information consumers to easily identify events in the past 24 hours. The volume and velocity of these streams causes information from prior days to quickly vanish despite its utility in providing an informative context for interpreting new information. Few capabilities exist to support an individual attempting to identify or understand trends and changes from streaming information over time. The burden of retaining prior information and integrating with the new is left to the skills, determination, and discipline of each individual. In this paper we present a visual analytics system for linking essential content from information streams over time into dynamic stories that develop and change over multiple days. We describe particular challenges to the analysis of streaming information and present a fundamental visual representation for showing story change and evolution over time.

Scalable Visual Analytics of Massive Textual Datasets

This paper describes the first scalable implementation of a text processing engine used in visual analytics tools. These tools aid information analysts in interacting with and understanding large textual information content through visual interfaces. By developing a parallel implementation of the text processing engine, we enabled visual analytics tools to exploit cluster architectures and handle massive datasets. The paper describes key elements of our parallelization approach and demonstrates virtually linear scaling when processing multi-gigabyte data sets such as Pubmed. This approach enables interactive analysis of large datasets beyond capabilities of existing state-of-the art visual analytics tools.

Enabling massive scale document transformation for the semantic web: the universal parsing agent ™

The Universal Parsing Agent (UPA) is a document analysis and transformation program that supports massive scale conversion of information into forms suitable for the semantic web. UPA provides reusable tools to analyze text documents; identify and extract important information elements; enhance text with semantically descriptive tags; and output the information that is needed in the format and structure that is needed.

Threat stream data generator: creating the known unknowns for test and evaluation of visual analytics tools

We present the Threat Stream Data Generator, an approach and tool for creating synthetic data sets for the test and evaluation of visual analytics tools and environments. We have focused on working with information analysts to understand the characteristics of threat data, to develop scenarios that will allow us to define data sets with known ground truth, to define a process of mapping threat elements in a scenario to expressions in data, and creating a software system to generate the data. We are also developing approaches to evaluating our data sets considering characteristics such as threat subtlety and appropriateness of data for the software to be examined.

Visualizing data streams

We introduce two dynamic visualization techniques using multi-dimensional scaling to analyze transient data streams such as newswires and remote sens- ing imagery. While the time-sensitive nature of these data streams requires immediate attention in many applications, the unpredictable and unbounded characteristics of this information can potentially overwhelm many scaling al- gorithms that require a full re-computation for every update. We present an adaptive visualization technique based on data stratification to ingest stream information adaptively when influx rate exceeds processing rate. We also de- scribe an incremental visualization technique based on data fusion to project new information directly onto a visualization subspace spanned by the singular vectors of the previously processed neighboring data. The ultimate goal is to leverage the value of legacy and new information and minimize re-processing of the entire dataset in full resolution. We demonstrate these dynamic visuali- zation results using a newswire corpus, a remote sensing imagery sequence, and a hydroclimate dataset.