SDM: Selective Dynamic Manipulation of Visualizations

Mei C. Chuah, Steven F. Roth, Joe Mattis, John Kolojejchick

School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213, USA
Tel: +1-412-268-2145
E-mail: mei+@cs.cmu.edu, roth@cs.cmu.edu


ABSTRACT

In this paper we present a new set of interactive techniques for 2D and 3D visualizations. This set of techniques is called SDM (Selective Dynamic Manipulation). Selective, indicating our goal for providing a high degree of user control in selecting an object set, in selecting interactive techniques and the properties they affect, and in the degree to which a user action affects the visualization. Dynamic, indicating that the interactions all occur in real-time and that interactive animation is used to provide better contextual information to users in response to an action or operation. Manipulation, indicating the types of interactions we provide, where users can directly move objects and transform their appearance to perform different tasks. While many other approaches only provide interactive techniques in isolation, SDM supports a suite of techniques which users can combine to solve a wide variety of problems.


KEYWORDS: Interactive techniques, visualizations, direct manipulation


INTRODUCTION

A well-designed visualization can be tremendously helpful in analyzing data. However, current static visualizations are limited in several important ways:

  1. Users are not able to focus on different object sets in detail while still keeping them in context with the environment. This is especially important in large data sets that have too much information to be displayed in detail at once.
  2. When the information space is dense, there will be a lot of clutter and object occlusion. Occlusion is a serious problem because data-points may be fully occluded and thus appear invisible to the user. In this way, occlusion may distort a user's perception of the data-space. In addition, clutter makes it difficult to perceive patterns in subsets of objects.
  3. A data set may contain elements that have vastly different values. Thus, some objects may be dwarfed when shown in the scale used for the entire data set. In Figure 1, many of the objects (e.g., those in green) are dwarfed by the tall cylinder towards the right. With SDM we attempt to provide users with techniques that temporarily change the scale of a subset of objects, while keeping them in context with the environment. Figure 2 shows the same data set as Figure 1 except that the green objects have their heights scaled differently from the rest of the environment. The difference in height scales is shown by the ratio axes on the left of the visualization. Note that some of the bars that appeared to have the same height in Figure 1 actually have relatively large height differences when put on another scale.


    Figure 1
    : Data visualization of a relief effort data set. Green bars are the currently selected objects.


    Figure 2
    : Visualization showing the same data set as Figure 1, except the heights of the green bars are scaled differently from the rest of the environment.

  4. Many visualizations only allow users to view the underlying data. No tools are provided to classify sets of objects and save those classifications as new information is discovered. For example, after a user has identified several problematic/important data points, the user might want to save them as a set and change their color so that they are differentiated from the other objects.
  5. It is difficult to compare quantities represented by graphical objects which are not spatially contiguous. For example, in Figure 3 it is difficult to compare which of the selected shelters (indicated by the green bars) has the most supplies (indicated by their heights) because they are not at the same distance from the user.


Figure 3
: Selected objects that are at different distances from the user

In this paper we present a paradigm, Selective Dynamic Manipulation (SDM), which deals with all of the difficulties mentioned above. While some of the individual techniques that we discuss have been tried before, they have only been tried in isolation. Instead of isolated solutions, our approach is to provide a suite of interactive techniques that can be flexibly combined to solve a wide spectrum of problems. We believe that the unifying framework behind the techniques will make the system easy to learn, and the flexibility in using the methods will stimulate users to come up with multiple novel ways to solve tasks.

Our approach is different from many other interactive systems [4,9,15,18] because its selection and control methods are strongly centered on objects rather than on spaces. The goal of our approach is to provide a framework for solving a basic set of tasks and problems that are important when users are analyzing large, diverse data sets.


EXAMPLE DATA DOMAIN

The data domain for our examples is from a supply distribution network for a relief effort in a large-scale crisis. In Figure 1, supply centers are represented by cylinders, main routes between them by dark lines on the floor plane, and shelters where supplies are needed by rectangular bars. The heights of cylinders and bars indicate the quantities of material available at supply centers and needed by shelters. See [14] for color versions of figures.

Next, we outline the basic structure and approach of SDM and discuss how SDM can be used to address the five limitations mentioned previously.


SDM COMPONENTS

The structure of our SDM system is described based on the three primary components of interactive techniques: method of selection, interactive operations, and the feedback mechanisms and constraints placed on the behavior and appearance of objects.

1) Object-centered Selection

In object-centered selection the focus or selected set is made up of objects. Selection of these objects is done through their visualized representations or through data constraints.

Selection in SDM can be accomplished by clicking on objects and/or using sliders to place data constraints. The slider interface in SDM allows users to combine data constraints both conjunctive and disjunctively. In the future, we plan to generalize the selection facility so that there can be multiple linked displays. Selecting objects in a display will cause objects that have the same underlying data, in the other displays, to be selected as well.

After object sets are created, they may be named and saved. Users may save as many sets as they desire. Sets that are created will appear in a scrollable menu. Selecting a set in the menu causes it to be displayed. All the slider panels that form the set will also be shown. In this way users may switch between multiple self-defined sets easily. Note also that the object sets need not be made up of homogeneous types (e.g., a set may contain supply centers, shelters and routes).

2) Dynamic and Flexible Operations

Users may directly manipulate object set parameters through object handles. Direct manipulation refers to operating directly on objects instead of through menus or dialogues. The Document Lens [12] and Magic Lens [18] are examples of direct manipulation techniques that allow users to control their views of objects by operating on a lens. In contrast to manipulating an external object (such as a lens), SDM allows users to control an object set by directly controlling any element within that set through handles. Figure 4 shows the SDM handles used to manipulate objects. Our use of handles was motivated by related work on 3D widgets [7,17] and by 2D scaling handles in draw programs.

Attaching a handle to an object in a selected set and pulling or pushing its parts causes some or all objects in the selected set to change. The handles (Figure 4) were designed to look similar for different object types so that users need not remember the functions of many visually distinct handles. When handles are pulled or pushed, the objects contract, expand or move continuously. Using animation in this way helps users perceive the changes that have occurred or are occurring to an object set more easily.


Figure 4
: SDM handles

SDM operators can be applied flexibly. SDM allows users to control object parameters singly or alternatively to control multiple parameters simultaneously. The latter can be achieved by linking the object controls or by linking the parameters through formulas. This is in contrast to more constrained interactions offered by fisheye lenses [4] or stretching [15], which allow only a fixed technique for distortion.

Through handles, users may select which object parameters to change. For example, in Figure 4 part (a) controls the radius of the cylindrical object, part (b) controls the height of the object, and part (c) controls the width of the bar. The arrow handles on top of each object in Figure 4 enable users to shift the objects anywhere in the 3D space. Parameters can also be combined so that they can be manipulated through one handle.

3) Object Constraints and Feedback Techniques

SDM is an attempt at introducing a new physics of objects that supports great malleability and flexible control. In this new physics, an object may pass through other objects in the scene or have its position or appearance altered. For example, objects may be elevated (Figure 9) or translated on the floor plane (Figure 7). Objects may also have their widths (Figures 6, 11, and 12) or heights (Figure 2) changed.

We believe that flexibility is key to the success of interactive techniques, but may lead users to lose context of the environment as well as misinterpret the data. Thus we have developed a small set of constraints and a suite of feedback techniques that we believe will help reduce confusion while still maintaining a high degree of flexibility.

Context persistence is one of the most important constraints in SDM. This is because many data analysis tasks require users to focus on or manipulate a select set of objects in the visualization, while still maintaining some relationship between the focus objects and the rest of the environment. For this reason we have constrained SDM operations to always maintain various degrees of context between the focus set and its environment. To this end we have also introduced feedback mechanisms that help users relate the focus set to other objects in the visualization without any confusion.

Another constraint in SDM is set-wide operations. All objects in the selected set are subject to the same forces applied to any object in the set. This is useful because it helps base SDM techniques on several real-world analogies. SDM move-shift operations are analogous to putting all objects in the selected set on a platform and then moving the platform. SDM width and height scaling operations are analogous to stretching a set of malleable objects from the top or side with the same force. All these operations maintain the relative position, width, and height of all objects in the selected set.

A defining feature of SDM is that changes are made to the objects and not to the view on the objects, as is done in [18]. Because of this, it is important to clearly indicate which objects have been altered, how they have been altered and how they can be returned to their home positions. To serve these purposes, SDM has a set of feedback mechanisms to reduce user confusion and increase system usability. These feedback mechanisms fulfill the following functions:

Clearly Identifies the Selected Set, and thus the Objects that will Change in Response to a User Action. This is done by painting the selected set differently from all other objects in the scene. In addition, a white grid may be drawn beneath all selected objects (as is shown in Figures 7 and 9) so that the approximate positions and spread of its elements can be easily identified.

Maintains Scene Context. When objects are displaced, users need to get feedback on the original object positions in order to maintain context with the rest of the environment. One way that SDM achieves this is by having multiple representations of an object. Each data object is represented by two graphical representations: the 'body' and the 'shell' graphical objects. Object shells are left behind in the home position when the object bodies are drawn out and displaced. Object shells always appear in the original object width and height. Figures 7, 9, and 13 show some examples of object sets which have been displaced and their shells, left behind at the original positions, widths and heights.

Maintains Temporal Continuity. Interactive animation is used to provide users with temporal continuity so that they can perceive, track and undo the effects of an operation on the scene. Animation helps users perceive changes to the scene without having to cognitively reassimilate relationships between the pre-action and post-action scenes [13].

This shift of cognitive load to the perceptual system allows the user to concentrate more on the results of an action rather than on the process of the action. It also provides additional feedback to the user as to which objects have changed and how they have changed relative to their previous states. Techniques such as fisheye lenses [4] do not have object continuity (interactive continuity). Focusing on one point causes a sudden magnification of that point and demagnification of other regions. Because of the abrupt change in the visualization it is easy for users to lose track of where they were before and how they got to the current state. User tests in [8] showed that insufficient temporal continuity caused poor user performance in some tasks.

Maintains the Relationships Between the Selected Set and the Environment. It is important to maintain the relationship between the selected set and other objects in the visualization so that elements within the selected set, which may have been altered, can be correctly interpreted with the rest of the environment. An example of such a feedback technique is the ratio axes technique (left of Figure 1 and 2). The ratio axes display the scale of the environment and the scale of the current selected set. By comparing the lengths of the axes, users can tell what the scale difference is between the selected set and its environment.

Allows Objects to be Easily Returned to Their Home Positions. Home mechanisms are available to return objects back to their home positions, as well as to their original width, height and color. In addition, we have implemented "object bumpers". It may be hard to manually move objects back to their home positions because such an action requires precise movement. With manual control, it is easy to accidentally overshoot the home point. For this reason, we provide object bumpers, where the initial object move indicates the direction of freedom, and from then on objects can only be moved back as far as their home positions.


APPLYING SDM TO VISUALIZATION PROBLEMS

We have discussed five useful operations in the data analysis process that are lacking in current static visualizations. Next we outline how SDM techniques can be used to fulfill those needs.

Focusing on a Select Set of Objects while Keeping Scene Context.

While navigating through an information space, it is often necessary for users to focus on certain parts of that space. However, during and after focusing, it is important for users to be able to tell where they are in the information space.


Figure 5
: Painting focus objects


Figure 6
: Expanding the selected set in place


Figure 7
: Pulling focus objects to the front

In SDM, focus can be achieved with several strategies. Objects may be painted so that they appear visually distinct from other elements of the environment. Objects can also be made more salient by increasing their widths. Figure 5 shows the original visualization and the selected object set (shown in green). Figure 6 shows the visualization after objects in the selected set have been expanded. Spatial context is maintained in these cases because the objects are not moved from their original positions. However, some context may be lost because expanding the selected objects may cause other objects in the scene to be occluded.

Alternatively, users may shift/move all focus objects to the front of the scene (shown in Figure 7). Context is still maintained because object shells (shown in white) are left behind. In addition, a user can move the focus set back and forth between its home position and its position at the front of the scene to achieve better scene context.

Viewing and Analyzing Occluded Objects.

Occlusion is closely related to the task of focusing while maintaining context. As was previously mentioned, expanding the selected objects may introduce occlusion, but this can be prevented by elevating the objects before expanding them.


Figure 8
: Initial view of occluded objects


Figure 9
: Selected object set is elevated to solve occlusion problem

Figure 8 shows a dense information space with the interest objects (barely visible) shown in green. Many of the interest objects are partially if not fully occluded. After elevating the interest objects we can clearly see the pattern that they form (Figure 9). Another way to deal with occlusion is to make all objects, other than the interest set, invisible. This method, however, sacrifices scene context.


Figure 10
: Objects other than the selected set are scaled to have zero height


Figure 11
: Objects other than the selected set are scaled to be very thin

The occlusion problem can also be solved by reducing to zero, the heights of objects that are not of interest (Figure 10). In this way, the spatial position of the other objects with respect to the interest objects is maintained. However, their height patterns are lost. Alternatively, users may make all objects, except the interest objects, very thin (Figure 11). This allows us to view the interest set clearly as well as use object heights to maintain information about the amount of supplies a center has or that a shelter needs. We are also exploring the use of transparency to deal with object occlusion.

Viewing Different Sets of Elements Based on Different Scales.

It is often the case, especially in large information spaces, that the widths and heights of objects are quite diverse. In these instances, the overall scale used may dwarf some objects, and it would be very difficult to observe width and height relationships among those dwarfed objects. This problem is worsened by the aggregation operation, which can create aggregate objects that show the total value of all its members. These objects will have much greater values associated with them compared to the other objects in the environment.

For these reasons it is important to provide ways to view different data sets at varying scales [6]. SDM allows users to do this. Suppose a user wants to compare a set of shelters (shown in Figure 1 in green) with low demand and see whether the demand distribution of those shelters are in any way related to the neighboring shelters, routes, and supply centers. To perform this task, the user can scale up the heights of the selected objects, so they can be compared more easily (Figure 2). The ratio axes on the left indicate that the selected objects are now eight times larger than the scale of the environment.

Interactively Augmenting the Visualization with New Classification Information.

It is often the case that a lot of new insights and information is revealed during data analysis. However, because this new information is not part of the underlying data, most visualization systems will not be able to visualize it until it is fully described in the database. This means that the newly discovered property will have to be fully characterized and values must be assigned for it to all elements. This is very cumbersome because it is often the case that users only have a partial characterization of the property and may only care about showing it for a subset of objects in the data set. It is therefore useful to be able to interactively classify subsets of objects in a visualization as new information is discovered about them.

Object classification is one of the main tasks in data analysis [16]. To support this task, SDM allows users to interactively change any property of an object or object set in the visualization so that they share some common properties that are not present in other objects in the environment. With this technique users can construct their own grouping/classification structure by making different subsets of objects distinguishable from each other and from objects in the environment. Note that the object sets may be changed and redefined at any time, so users are free to change the definitions of the classification structure any time throughout the data analysis process.

In Figure 12 the user has identified some important shelters. The user has chosen to increase the widths of these shelters as well as color them yellow, to differentiate them from the other shelters. The new features that have been added, width and color, only make sense to the user or group of users that are familiar with this convention. Nevertheless, this technique is useful because it offers a quick way to add information into the visualization, very much like the way we annotate printed text or paper maps.


Figure 12
: Visualization augmented to show a new property, danger areas.

Comparing the Patterns, Widths, and Heights of Objects.

In the data analysis process, users frequently compare the various attributes of data objects. In a visualization, these attribute values are represented by graphical objects; thus this task reduces to comparing the different parameters of graphical objects. However, comparisons of object widths and heights are especially difficult to make in 3D visualizations. This is because it is difficult to compare the widths or heights of objects that are at different distances from the user (e.g., the green bars in Figure 3). The same problems are encountered in 2D visualizations when trying to compare the lengths of interval bars.


Figure 13
: Two sets of objects made for height comparisons

SDM allows users to easily perform comparisons among objects of different depths by giving users the ability to draw a line of reference in the scene plane. Users can then move any set of objects to the reference line. By lining up objects in this way, their heights can be easily compared. Figure 3 shows the initial visualization and the selected objects (in green). Figure 13 shows the visualization after the selected objects have been pulled to the reference line. To maintain the relationship between objects and their shells, users may paint the objects which will cause their shells to also be painted. Users may also slide the objects back and forth between the reference line and their home points.

In addition to making comparisons within sets, users may also compare height trends among multiple sets. Figure 13 shows that the user has lined-up two different sets of objects. From the two lined up sets it is clear that one set (in green) has larger height variations than the other.

Another typical comparison operation is to find patterns formed by different object sets and then look for property differences among those patterns. In large data spaces, it is difficult enough to notice patterns among objects, much less compare them. This problem can be solved in SDM by separating out the subsets with the shift-move operator. The spatial shift operation maintains relative positions among elements within each subset; thus the patterns they form are preserved. The subsets may be further manipulated in their separate spaces to tease out any important differences or similarities.

In this section we have presented some techniques that can be used to aid users in exploring and analyzing large data sets. However, this set of techniques is by no means comprehensive. We envision that as users learn and get more comfortable with the system, they will start to combine the basic operations in new ways to fit their preferences in solving the task at hand.


IMPLEMENTATION

The current SDM implementation was developed in GL and C on a Silicon Graphics Indigo workstation. Figure 14 shows the general system architecture. The SDM architecture can be divided into three main sections: the data modeling section, the physicalization section and the manipulation section. We discuss each of these sections in greater detail below.

Data Modeling

The main component in the data modeling process is the data set builder. The data set builder takes data objects and groups them into data sets, based on the task being performed. A data set denotes a group of objects that will be visualized in a uniform manner, and as such can only contain homogeneous data object types (e.g., a data set of supply centers cannot also contain supply routes). However, a single data object can occur in multiple data sets. This allows, for example, a supply center to be grouped into both the "Critically Important" data set and the "Low on Supplies" data set.


Figure 14
: SDM system architecture and information flows

Physicalization

The physicalization process is based on the metaphor of creating "physical" objects in order to represent abstract data objects. The main component that performs this process is the instantiator, which has the responsibility of taking a selected data set and visualizing it as one or more graphic sets. Every object in a graphic set corresponds to a unique object in the data set that it represents.

Each object in a graphic set utilizes the same visual specifications. The specifications define the class of graphic object (e.g., cylinder, bar, sphere, etc.) that the graphic set will contain, along with the mapping between select attributes of the data objects and the various physical properties of the graphical objects. For example, a data set of supply centers might be visualized as a set of cylinders; where the "materials-weight" attribute is mapped to the height of the cylinder, and the "longitude" and "latitude" attributes are mapped to the x and y location of the cylinder.

Encoders perform the actual conversion of data-values (e.g., pounds of food) into graphical-values (e.g., height in inches). Some examples of encoders are x-axis encoders, y-axis encoders, and color-key encoders. Physical object properties that do not encode a data attribute are assigned a default value.

A single data set may be visualized as multiple independent graphic sets, in single or multiple data-graphics. Objects from the various graphic sets may encode different attributes of the data set, or may encode identical attributes of the data set multiple times. Allowing a data set to be multiply represented is useful in the case where a single graphical set is insufficient to effectively show all important attributes of the data set. It is also useful in those cases where we want to strengthen the effect of a data attribute by encoding it twice.

Manipulation

Once data objects have been "physicalized" as graphic objects, SDM gives users the ability to move and stretch the graphic objects, analogous to the manner in which users might manipulate physical models placed on a table. However, the "physics" provided by SDM is not limited to real-world manipulations; users can also elevate, compress, and perform other operations upon objects that wouldn't be possible with actual physical models. Thus, SDM exploits common physical metaphors so that users may quickly get comfortable with the system, and then extends these metaphors to take full advantage of the capabilities that real-time computer graphics provide.

There are two main components to the manipulation process: the selection manager and interactors. The selection manager enables users to group graphical objects into alternative focus sets. Users can group objects by selecting them with the mouse or by manipulating sliders and buttons, actions which the selection manager interprets to construct a focus set. Users can switch between focus sets in order to quickly locate objects that are relevant to the current task. Users may also save any focus set that might be relevant to a later task. This focus set will be present the next time the system is invoked.

The other major component of the manipulation is the set of interactors. Each focus set is associated with an interactor, which is responsible for changing the physical appearance of graphical objects in that set. Interactors are controllable through the handles that can be attached to objects in the visualization. Pushing or pulling on a handle causes the interactor to transform the values of select properties of graphical objects. Thus, encoders generate the original appearance of the graphical objects, based on the data that they represent, while interactors perform transformations on these properties to manipulate the objects as serves the user's purpose.

Values are only changed for those objects which have the properties that are being controlled. Because SDM allows for heterogeneous focus sets, it is possible that a set may contain bars, cylinders and lines. Operating on the radius property only affects the cylinders, since radius is not a property of bars or lines.

An interactor may be classified based on its two properties: cardinality and type. Cardinality indicates the number of properties that the interactor affects, and may be single or multiple. Single property interactors include those that change the height of an object, and those that move the x-position of an object. Multiple property interactors use constraints or evaluation functions to determine a "derived" property for an object, based on several of its basic properties. An example of a multiple property interactor is the alignment interactor, which transforms object positions in order to scale their perpendicular distance to a line on the object plane.

Type indicates the nature of the graphical transformation, and may be scale or shift. Shift interactors add constants to object properties. They are usually applied to graphical properties that are not relevant to the user's immediate task in order to make other properties more noticeable. An exception to this is position, since shifting objects to the front can make their relative positions more apparent. Scale interactors, on the other hand, proportionately transform property values so that they maintain their ratios relative to each other. They are typically used to enhance differences among the data values that are encoded.


PREVIOUS WORK

Previously, we discussed how SDM deals with the five limitations of current static visualizations. In this section we discuss how other interactive systems deal with these limitations. In doing this we highlight some of the major differences and the advantages and disadvantages of SDM compared to other systems.

Focusing on a Select Set of Objects while Keeping Scene Context.

Focus areas can be made larger in two ways: object magnification and spatial magnification. SDM uses object magnification, which only acts on the objects and not on the space between objects.

Magnification in other interactive techniques [3,4,15,18] is spatial because expanding the selected set causes the entire selected space to expand and not just the objects within it. Thus, the objects move out of their original positions and farther away from each other as their size increases. In [8], subjects found the change in distances between objects to be disorienting and jarring. In addition, because the between-object spaces have expanded, either some other space has contracted (as in [4] and [15]), or some objects are now occluded by the newly expanded region (as in a magnification magic lens [18]). Another disadvantage of spatial magnification is that is works well on tasks that are strongly based on object spatial position but fail when users want to focus on multiple object sets based on non-spatial properties.

Like some spatial magnification techniques, object magnification also faces the problem of occlusion. In object magnification many of the newly expanded objects may occlude or collide with other objects. This problem can be solved in SDM by elevating the expanded objects.

In contrast to spatial magnification, however, object magnification works for both spatially and non-spatially based tasks. For spatially based tasks users can simply collect all the data points in the desired space (either through their longitude and latitude or by selecting them from the visualization) and form a data set. Operating on that data set would effectively produce the effect of operating on the area that contains the data points. One disadvantage of object magnification, however, is that users cannot operate on parts of objects. For example, enlarging a particular road would mean that the whole road has to be enlarged and not just part of it. To solve this problem we are currently exploring techniques that combine spatial and object metaphors.

Some interactive techniques do not use magnification for focusing. Instead, they allow users to focus on certain data points (objects) by giving users the ability to change the appearance of object groups in real-time. Objects that are not in focus are either temporarily invisible [1] or uncolored [11] (i.e., they have a different appearance than the objects that are in focus). Users may easily regain context by making the invisible points reappear [1] or from the uncolored points [11].

Another important issue of focusing in context is the scope of the operation. Techniques such as fisheye lenses [4] and stretching [15] have unlimited scope because changes to the selected set causes changes to occur throughout the visualization. SDM and other techniques such as magic lenses [18] have limited scope in that operations affect only the selected space(s). Unlimited scope may be disconcerting to the user because a local action may affect objects in the entire visualization. A disadvantage of limited scope is that it may cause certain objects/regions to be occluded by the focus objects/regions.

Viewing and Analyzing Occluded Objects.

In dense information spaces, object occlusion is a major problem. Dynamic query sliders [1] solve this problem by providing users with the ability to flexibly change the visibility of objects. Other techniques deal with occlusion by only partially displaying the objects. For example in [2], when the telephone links are too dense, only sections of the links are shown to reduce occlusion. A problem with the approach in [2] is that the technique is limited to the length of the lines. In SDM, we propose more general techniques that apply to multiple object types.

Viewing Different Sets of Elements Based on Different Scales.

Under the stretching paradigm [15], this problem may be solved by pulling on the region surrounding each of the small objects in order to extend their heights. Care must be taken that the force applied to all objects is equal so that the object heights can be fairly compared. To solve this problem using fisheye lenses would be just as awkward because all dimensions of the objects would be magnified, not just its heights.

Another way to analyze objects that have widely differing widths and heights is to change the width and/or height scales to a logarithmic scale. In this way, width/height order is maintained. However, the linear relationships between the objects are lost. Therefore, it would be hard to compare the heights of two or more bars and determine approximately how much wider or taller one is than another.

Interactively Augmenting the Visualization with New Classification Information.

Painting is a way to classify object sets by using color. SDM allows users to classify objects not only by changing their color but also by changing any other parameter. In addition, current classification methods only have temporary effects, whereas SDM allows users to save objects sets to disk and restore them in future sessions.

Comparing the Patterns, Widths and Heights of Objects.

The calendar visualizer [10] solves the problem of width and height comparisons by using lighting to cast shadows on the back and side planes of the visualization (i.e., projecting an orthogonal image/shadow of the scene onto a flat plane). The main problem with this method is difficulty in identifying an object and its corresponding shadow. In addition, if the information space is dense, there will be too much overlap between shadows of the focus set and shadows of objects that are not relevant.

In general, the approach taken by SDM is different from other approaches because SDM provides a flexible set of methods that allow users to manipulate their information space in many different ways. Many other interactive techniques only allow users to distort the information space based on a fixed method. Fisheye lenses [4] and stretching [15] allow users to increase the area of the focus region by reducing the area of surrounding regions. Painting [11] allow users to focus on the selected set by changing its color.

An exception to this is the magic lens technique [18], which allows many different distortions depending on the lens that is being used. Like SDM, that research defines a unifying paradigm, derives a tool set from it, and provides composition of multiple operations. SDM operations, however, are designed to enable users to directly manipulate object appearance as they would in a draw program. Lenses, on the other hand, provide metaphors for viewing regions containing objects in different perspectives and can be general-purpose (e.g., to distinguish occluded objects) or application-specific (e.g., to view semantically related properties of objects). Unlike SDM, lenses do not address parameter manipulation per se.


CONCLUSION AND FUTURE WORK

We have presented a suite of techniques based on the SDM paradigm. These techniques enable users to:

Many current interactive techniques tend to examine and present solutions to isolated problems. The goal of SDM is to provide users with enough tools and flexibility that they can solve a wide spectrum of data analysis tasks. The flexibility provided also presents users with multiple alternative solutions for any given task. Although the examples presented in this paper use rectangular objects and cylinders, the same techniques apply just as well to other graphical objects.

In the future, we plan to integrate current SDM operations with other aggregation and interactive data manipulation interface techniques [5, 14]. We also intend to explore the possibility of using SDM to perform "what if" analyses, where changes to objects in the visualization actually propagate to the data objects that they represent.


ACKNOWLEDGMENTS

Funding for this project was provided by the Advanced Projects Agency (DAAA1593K0005) and the Army Research Laboratory.


REFERENCES

1. Ahlberg, C., Williamson, C., and Shneiderman, B. Dynamic queries for information exploration: An implementation and evaluation. Proceedings CHI `92 Human Factors in Computing Systems, ACM, May 1992, pp. 619-626.

2. Becker, R.A., Eicks S.G., and Wilks, A.R. Visualizing network data. IEEE Transactions on Visualization and Graphics, March 1995, pp. 619-626.

3. Bederson, B.B., and Hollan, J.D. PAD++: A zooming graphical interface for exploring alternate interface physics. UIST `94, November 1994, pp. 17-27.

4. Furnas, G.W. Generalized fisheye views. Proceedings CHI `91 Human Factors in Computing Systems, ACM, April 1991, pp. 16-23.

5. Goldstein, J., Roth, S.F., Kolojejchick, J., and Mattis, J. A Framework for knowledge-based interactive data exploration. Journal of Visual Languages and Computing, No. 5, 1994, pp. 339-363.

6. Herndon, K.P., van Dam, A., and Gleicher, M. Workshop report: The challenges of 3D interaction. CHI '94 Human Factors in Computing Systems, ACM, May 1994, pp. 469.

7. Herndon, K.P., and Meyer, T. 3D widgets for exploratory scientific visualization. UIST `94, November 1994, pp. 69-70.

8. Hollands, J.G., Carey, T.T., Matthews, M.L., and McCann, C.A. Presenting a graphical network: A comparison of performance using fisheye and scrolling views. Designing and Using Human-Computer Interfaces and Knowledge Based Systems, Elsevier Science B.V., Amsterdam, 1989, pp. 313-320.

9. Lamping, J., Rao, R., and Pirolli, P. A focus+context technique based on hyperbolic geometry for visualizing large hierarchies. Proceedings CHI '95 Human Factors in Computing Systems, ACM, May 1995, pp. 401-408.

10. Mackinlay, J.D., Robertson G.G., and DeLine, R. Developing calendar visualizers for the information visualizer. UIST '94, November 1994, pp. 109-119.

11. McDonald, J.A. Painting multiple views of complex objects. ECOOP/OOPSLA '90 Proceedings, October 1990, pp. 245-257.

12. Robertson, G.G., and Mackinlay J.D. The document lens. UIST '93, November 1993, pp. 101-108.

13. Robertson, G.G., Mackinlay, J.D., and Card, S.K. Cone trees: Animated 3D visualizations of hierarchical information. Proceedings CHI '91 Human Factors in Computing Systems, ACM, April 1991, pp. 173-179.

14. Roth, S.F. The SAGE Project.
http://www.cs.cmu.edu/Web/Groups/sage/sage.html

15. Sarkar, M., Snibbe, S.S. Stretching the rubber sheet: A metaphor for viewing large layouts on small screens. UIST `93, November 1993, pp. 81-91.

16. Springmeyer, R.R., Blattner, M.M., and Max, N. L. A characterization of the scientific data analysis process. Visualization '92, October 1992, pp. 235-242.

17. Stevens, M.P., Zeleznik, R.C., and Hughs, J.F. An architecture for an extensible 3D interface toolkit. UIST `94, November 1994, pp. 59-67.

18. Stone, M.C., Fishkin, K., and Bier, E.A. The movable filter as a user interface tool. Proceedings CHI '94 Human Factors in Computing Systems, ACM, April 1994, pp. 306-312.




Return to SAGE Home Page

Sean Cier (scier@cmu.edu)
Last update: 5 December 1995