...paper, space-time-attribute refers geographically referenced, time-varying data involving multiple thematic attributes; the focus of methods and tools described is on identifying and interpreting spatio-temporal, multivariate patterns in these data. Existing approaches to pattern identification and interpretation, from entirely computational to visually led methods, are limited in analyzing complex patterns that include space, time, and attribute components together. Moreover, traditional information visualization methods do not support analysis of large data sets. Pattern recognition, machine learning, and other computational methods have been developed explicitly to deal with large and high-dimensional data sets, but typically do not provide ways to incorporate both space and time, nor do they leverage the power of human vision and cognition to help analysts notice and quickly interpret patterns in complex data. The goal of this research is to bridge this gap by developing analytic methods that couple visual, computational methods and human expertise in productive ways. The approach presented here was developed within the broad research framework provided by visual analytics, defined as "the science of analytical reasoning facilitated by interactive visual interfaces" (Thomas and Cook 2005, p. 4).

This research introduces a Visual Inquiry Toolkit (VIT) which provides information analysts with a flexible interface to integrated visual, computational, and cartographic methods that support an overview +detail strategy for identifying and interpreting patterns in space-time-attribute datasets of relatively large size. Overview +detail describes a strategy for supporting multiple levels of detail in an interactive visual display (Plaisant et al. 1995). This strategy is best known through Shneiderman's (1996) information-seeking mantra: overview first, zoom and filter, with details on demand. We propose adding a step to Shneiderman's mantra--information synthesis, which refers to capturing novel, relevant patterns and reorganize them to yield more useful information. Beyond support for the extended visual overview +detail strategy, the VIT also emphasizes flexible interaction strategies designed to enable human knowledge and judgment to be coupled productively with computational pattern-finding methods to support an iterative analysis process.

The remainder of the paper is organized as follows. In the next section, we review related literature. Following that, we discuss our strategy and methodologies, with a focus on representation issues; then, we demonstrate an interactive visual analytics approach for identifying and interpreting spatio-temporal multivariate patterns. Finally, the advantages and limitations of the approach and possible further work are discussed.

Related Work

A starting point for our approach is past work on visualizing multivariate data. The commonly used data representations for multivariate visualization include tables and scatter plots; more sophisticated methods include scatterplot matrices (Andrews 1972), parallel coordinate plots (Inselberg 1985), matrix permutation (Makinen and Siirtola 2000; Bertin 1981), and multivariate glyphs (Pickett et al. 1995). A comprehensive review of the methods can be found in a paper by Keim et al. (2005). All of these methods, however, have difficulty representing large data sets. As the number of data items/variables goes up, the potential for over-plotting on displays goes up as well. Two major solutions have been proposed to address this problem. One is to reduce the data size being displayed by grouping individual data records into subsets (e.g., aggregation or clustering); in this case, collective characteristics of the grouped data are visualized and investigated (Guo et al. 2005; Johansson et al. 2004; Ward 2004). The other solution is data selection, which allows zooming, filtering, and focusing on a subset of data (Keim et al. 2005). This research takes a combination of both approaches.

Visualization of space-time-attribute data are challenging because traditional single 2D or 3D views do not provide enough dimensional space to display all space, time, and multiple attribute components simultaneously. A widely adopted method for space-time data is to represent these data in a three-dimensional view where time data are visualized in the third dimension over a two-dimensional map (Kwan 2000; Lodha and Verma 2000; Kapler and Wright 2004). This method, however, has severe limitations for visualizing multivariate data of even modest size (e.g., hundreds of data records for more than two or three variables). Some other systems use animation to display time, presenting sequential representations of spatial information at a moment of time (Slocum et al. 2000; Oberholzer and Hurni 2000; Stojanovic et al. 1991). However, this technique imposes burdens on human short-term memory to retain temporal changes, thus it is not suitable for complex, large data sets. Two approaches that show some potential to address these issues are: (1) small multiple adjacent views (MacEachren et al. 2003) and (2) linked views (MacEachren et al. 1999; Andrienko and Andrienko 2001; Robinson et al. 2005). We extend both approaches, combining them with computational clustering methods.

Successful analysis of large, space-time-attribute datasets requires more than advances in visual representation or computational methods. Human interaction also plays important roles in identifying and interpreting complex patterns. Considerable effort has been directed toward methods for interactively detecting multivariate patterns (Harri 2004; Seo and Shneiderman 2002) and temporal patterns (Buono et al. 2005; Carlis and Konstan 1998). While several studies have focused on spatio-temporal data (Gatalsky et al. 2004; Kwan 2000), few approaches have been developed to interactively search for patterns using strategies that consider all aspects of space, time, and attribute components.

Most of the research within the geovisualization and information visualization communities on interaction has been focused primarily on developing methods and mechanisms to support real-time interaction with individual and linked views, using brushing, linking, focusing, and other direct manipulation methods (Shneiderman 1997; Andrienko and Andrienko 1999; Dykes 2005). As has been outlined from the perspective of both science (Gahegan 2005) and intelligence analysis (Pirolli and Card 2005), however, a goal of both exploratory geovisualization and visual analytics, generally, is to support an analytical process that is often complex, iterative, and carried out over an extended period of time. Thus, approaches to support interaction need to move beyond interaction as an action to interaction as a process. A key component in supporting an analytic process is provision of interaction methods that allow analysts to create, save, retrieve, and share analytic artifacts (Pike et al. 2007). The concept of a pattern basket, detailed below, is a step in this direction intended specifically to support saving, comparing, revising, and sharing patterns identified in complex space-time-attribute data sets.

Our own previous work (Guo et al. 2006) specifically addresses space, time, and multiple attributes. In that complementary work, the computational and visual methods are integrated and applied to single session analysis (e.g., an exploration session intended to uncover hidden patterns and/or generate hypotheses about multivariate relationships). The visual-computational tools described in the paper cited above, while interactive, put the emphasis on computational methods and offered relatively limited human interaction support and no explicit support for a sequential knowledge building process. The work introduced in this paper emphasizes support for a spiral multi-session analysis process with a systematic overview +detail strategy, allowing human judgment to steer the analysis process, refine the computational outcomes, and synthesize relevant, potentially useful information. Specifically, our methods and tools incorporate flexible human interaction focused on process, including: a highly manipulable, parallel coordinate plot that supports overview plus detail analysis; a dynamic dendrogram integrated with a reorderable matrix; and pattern baskets as a mechanism for supporting a multi-step analytical process of pattern identification and interpretation.

This paper expands in several ways upon a preliminary report on the above extensions presented in (Chen et al. 2006). For example, we detail the complementary roles played by various visual and interactive methods. We also clarify the way in which a Self-Organizing Map (SOM) can facilitate multivariate analysis and provide details on the color encoding applied to the multivariate clusters generated. Another instance of expansion is the provision of details on how to construct an holistic overview and detailed views for visualizing spatio-temporal, multivariate patterns, and how the techniques of static link and dynamic link, combined with the color scheme generated by the SOM, facilitates the construction of the overview. And, we formally introduce and illustrate application of the Pattern Basket concept--a reasoning artifact to facilitate externalizing cognition (for discussion of reasoning artifacts in visual analytics, see chapter 2 in Thomas and Cook, 2005); and clarify the concept of information synthesis and the way pattern baskets supports it.

Visual Inquiry Toolkit: An Integrated Approach

In this section, we provide a detailed, six-part introduction to the components of the Visual Inquiry Toolkit. First, we outline the tasks for which the toolkit is...

NOTE: All illustrations and photos have been removed from this article.



More articles from Cartography and Geographic Information Science
Humorous maps: explorations of an alternative cartography., January 01, 2008
Literature review.(Bibliography), January 01, 2008

Looking for additional articles?
Search our database of over 3 million articles.

Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication name or publication date.

About Goliath
Whether you're looking for sales prospects, competitive information, company analysis or best practices in managing your organization, Goliath can help you meet your business needs.

Our extensive business information databases empower business professionals with both the breadth and depth of credible, authoritative information they need to support their business goals. Whether it be strategic planning, sales prospecting, company research or defining management best practices - Goliath is your leading source for accurate information.