
Overview www.wiley.com/wires/compstats FIGURE 8| A screenshot of GeoTime. as our perception and cognition are concerned, the another. The proximity and similarity principles can whole is more than the sum of parts. Most commonly be used simultaneously to reinforce each other. known Gestalt principles are proximity, similarity, continuity, closure, figure and ground, and symmetry. The Mantra of Visual Information Seeking A comprehensive explanation of these principles and The most widely known visual information seeking a rich set of examples can be found in How Maps 41 mantra is given by Ben Shneiderman, University Work in the context of cartography, that is, the of Maryland: Overview first, zoom and filter, design of geographically based thematic maps. then details-on-demand.44 This mantra insightfully The proximity principle says that we tend to see summarizes the essential elements of interacting with groupings of individual items in a visual arrangement graphically presented information. based on the proximity between these items. Items Designers of visual overviews commonly capi- that are relatively close to one another tend to give us talize on metaphors that can give users a sense of a sense of similarity. In other words, we see individual intuitiveness and familiarity. Naturally, metaphors items in groups of some underlying similarity. This of an information space are particularly popular, principle has been adapted by the information visu- especially in 1990s, ranging from two-dimensional alization community from the early stage. Algorithms maps, three-dimensional landscape views and con- that can arrange information items in this fashion tours, to star fields and galaxies of information. An tap into the proximity principle. Some interesting important function of an overview is to depict inter- examples include Bead42 and Information Mural.43 relationships among units of information. Units of The similarity principle from Gestalt psychology textual information include words, sentences, docu- says that visual attributes such as the shape, color, and ments, and collections of documents such as websites. texture are cues for us to group items, for example, Units of visual information include scenes, episodes, all the circles in one group and all the triangles in and libraries of videos. 394 2010 John Wiley & Sons, Inc. Volume 2, July/August 2010 WIREs Computational Statistics Information visualization Information space metaphors naturally invite the context of scientific discovery in cognitive science, navigational operations such as zoom, pan, tilt, and psychology, and history of science. Few connections rotate. One of the earlier claims and design goals of have been established so far between the study of information visualization is that good information insight in other disciplines and the field of information visualization should present information to users intu- visualization. An intriguing introduction to some itively. Many filtering operations have been adapted of the recent understanding of the brain activities to enable users interact with dynamic information that lead to insights can be found in a New Yorker visualization, including brushing, linking, dynamic article The Eureka Hunt.45 The Nature of Insight is queries, and coordinated views. a comprehensive collection of studies of insight.46 In 1996, Shneiderman offered a taxonomy for In a recently developed explanatory and com- visual information seeking.44 The taxonomy divides putational theory of scientific discovery, the nature general visual information seeking into seven data of insight is characterized by a brokerage mech- types and seven tasks. This taxonomy is one of anism and a burst function of recognition.47 The the earliest and most influential contributions to the brokerage mechanism echoes what is described in information visualization field. the Eureka Hunt in that one arrives at insights by linking previously unconnected thoughts. The theory Seven Data Types is computational and it is possible to formulate the • one-dimensional data; search for insights as a problem of searching for the • potential linkage between even the most unthinkable two-dimensional data; relations. Initial studies of transformative discoveries • three-dimensional data; such as Nobel Prize winning discoveries are particu- • temporal data; larly promising. This approach is particularly relevant • multidimensional data; to visual analytics and insight-based evaluative stud- • ies because they can characterize insightful patterns in tree data; terms of structural and temporal properties. • network data. Within the information visualization commu- nity, notable efforts on characterizing and measuring Seven Tasks insights include exploratory approaches as opposed 9 • overview; to benchmark-based experimental studies, lessons learned from the first 3 years of InfoVis contests,48 • zoom; and more recent reflection in the context of visual • filter; analytics.8 An interesting framework of evaluating • details-on-demand; interactive visualizations is proposed recently in • relate; Ref 49. The framework is built on top of a generic conceptual model in human–computer interaction, • history; namely Don Norman’s Seven Stages of Action.50 • extract. According to the Seven Stages of Action, two stages of interacting with computer interfaces are particularly The data type by task taxonomy has influenced problematic: execution and evaluation. The gulf of a generation of information visualization researchers. execution and the gulf of evaluation are used to refer Other notable efforts include the data state reference to these problematic stages. The gulf of execution, model.3 for example, should be narrowed so that users can accomplish their tasks smoothly and seamlessly. The gulf of evaluation should be narrowed so that users The Pursuit of Insights can judge their progress accurately. Reflections on insight-centric evaluation are moti- Much of the discussions in information visu- vated by the increasing concern of how to establish alization on insights primarily address practical and the effectiveness of interacting with information visu- methodological issues concerning how evaluative alization interfaces. On the one hand, it is almost a studies should be designed to capture the effectiveness community-wide consensus that insight is the ultimate of an information visualization design in terms of goal of information visualization. On the other hand, insights. The types of insights that are relevant the definition of insight in the information visualiza- to information visualization and evaluative studies tion literature per se has been vague and ambiguous. have theoretical and practical implications. We The nature of insight has been extensively studied in found two meta-analysis studies of information Volume 2, July/August 2010 2010 John Wiley & Sons, Inc. 395 Overview www.wiley.com/wires/compstats visualization.51,52 Given the growing calls for level, scalability issues remain to be a long-lasting theoretical foundations in the field, this is expected to challenge.55 Some of the algorithms developed for be a significant topic of research. clustering large-scale data sets in machine learning are particularly appealing, such as Refs 56,57 and one Theoretical Frameworks can expect these algorithms will soon find their ways to information visualizations. The general consensus, as reported by a recent workshop and a few other public presentations, was that information visualization currently lacks Edge Bundling 53 adequate theoretical foundations. As a result, many Edge bundling is an emerging strategy to solve a com- approaches are ad hoc in nature. A week-long sem- mon problem in visualizing a densely connected graph inar took place at Dagstuhl, Germany in mid-2007, due to cluttered images caused by overlapping edges. for example, addressed four potential directions for Avoiding edge crossings has been long recognized as developing new theories. The lack of theories becomes one of the constraints that could improve the clarity of particularly prominent in information visualization resultant visualizations. Recently, an interesting strat- courses and when designing empirical and evaluative egy has emerged—that is the use of edge bundling tech- studies. niques in a variety of graph visualizations to increase The search for theoretical foundations increas- the clarity of visualized patterns. Bundling reduces ingly introduces and adopts theories and conceptual visual clutter. Visualizations with bundled edges make frameworks from other fields and disciplines. For it easier for viewers to see underlying patterns than example, distributed cognition in human–computer non-bundled versions,58 for example, as shown in Fig- interaction is seen as a potential candidate for a the- ure 9. Edge bundling is a generic technique in nature 54 oretical framework for information visualization. because it can be applied virtually to all node-and-link Norman’s Seven Stages of Action, also in human–com- diagrams regardless the underlying layout algorithms. puter interaction, provides a new insight into inter- In this sense, it is similar to other generic display tech- acting with information visualizations, specifically on niques such as fisheye views. A geometry-based edge 49 the gulf of execution and the gulf of evaluation. bundling example appears recently, showing promi- Many information visualizations lack a quanti- nent patterns of migration in the USA59 (Figure 10). tative measure that could indicate the overall
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages5 Page
-
File Size-