| Stuart Pook1,2,4 | Eric Lecolinet1 | Guy Vaysseix2,3 | Emmanuel Barillot2,3 |
|---|
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies
are not made or distributed for profit or commercial advantage and
that copies bear this notice and the full citation on the first page.
To copy otherwise, to republish, to post on servers or to redistribute
to lists, requires prior specific permission and/or a fee.
AVI 2000, Palermo, Italy
© 2000 ACM 1-58113-252-2/00/0005..$5.00
Zoomable User Interfaces (ZUIs) are difficult to use on large information spaces in part because they provide insufficient context. Even after a short period of navigation users no longer know where they are in the information space nor where to find the information they are looking for. We propose a temporary in-place context aid that helps users position themselves in ZUIs. This context layer is a transparent view of the context that is drawn over the users' focus of attention. A second temporary in-place aid is proposed that can be used to view already visited regions of the information space. This history layer is an overlapping transparent layer that adds a history mechanism to ZUIs. We complete these orientation aids with an additional window, a hierarchy tree, that shows users the structure of the information space and their current position within it. Context layers show users their position, history layers show them how they got there, and hierarchy trees show what information is available and where it is.
ZUIs, especially those that include these new orientation aids, are difficult to use with standard interaction techniques. They provide a large number of commands which must be used frequently and on a changing image. The mouse and its buttons cannot provide a rapid access to all these commands without new interaction techniques. We propose a new type of menu, a control menu, that facilitates the use of ZUIs and which we feel can also be useful in other types of applications.
ZUIs are used to present an information space to users. One of the reasons that users are unable to successfully use ZUIs is that the view of the information space shown to users, or the focus, does not always contain the context needed by users to position this focus in the information space. Once users are in this situation they are disoriented, sometimes to the point of not understanding what they are looking at and not knowing whether they should pan, zoom, or dezoom to find what they are looking for. It could be said that they are `lost in hyperspace.'
We present two new temporary aids that the system can display at the users' command if they arrive in this situation. The first, a context layer, allows users to position the focus with respect to more global views of the information space. The second, a history layer, allows users to revisit the route they took through the information space to arrive at their present position. We also present a third aid, a hierarchy tree, that is always visible. This aid is a second window in the ZUI and shows at all times the structure of the information and the users' current position within this structure. Users can also use this window to change their position in the information space.
ZUIs are complex programs with complex user interfaces that are normally controlled using the mouse, buttons and standard menus. When navigating in a ZUI, users zoom, dezoom, scroll, create magic lenses, move and resize magic lenses, move and scrolls portal, etc. Some of these actions are executed very frequently. A user zooms until the desired scale has been obtained and scrolls until the object looked for has been found. The graphical objects used to present the information space change frequently; a zoom or dezoom can completely change the objects visible. Making these objects active and using them to control the ZUI is problematic because these objects change and move too frequently. We propose a new type of menu, a control menu, that allows users to control ZUIs in a consistant and rapid fashion. A control menu can also include up to two scroll bars; a single interactor can thus control a complex operation.
A ZUI incorporating these new techniques can be tested over the Web at the URL http://www.infobiogen.fr/services/zomit/.
In contrast to the above methods that integrate the context and focus by deformation, we integrate the context and the focus by drawing the context (the context layer) over the focus. These two views are transparent so that they can both been seen at the same time. User studies [8] have shown that transparent views and overlays are well accepted by users.
When using a context layer two orthogonal controls are available: the scale of the view shown in the layer and the relative levels of transparency of the context layer and the focal view. The scale of the context layer can be chosen so that the context layer shows any view between the initial view and the focus. Users can also control the relative transparencies of the context layer and the focus. This allows users to concentrate on either the context or the focus by making the chosen view be drawn solid and the other as transparent as desired. The interactor used to control the scale and transparency is described in section 5.
The context layer is positioned so that the region of this layer that corresponds to the focus is in the centre of the ZUI's window. This region is indicated by a rectangle drawn in the centre of the window that shows the size and position of the focus relative to the view currently visible in the context layer.
This method of combining the focus and the context avoids the deformation of the other methods described above, which often makes images difficult to recognize and understand. The advantage of deformation, that more information can be represented, is maintained by a very fluid and rapid control of the scale of the context layer. The users can quickly find the context that they need to identify the focus. Changing the scale of the context layer changes the position and often the form of the objects in the layer. This movement helps users understand which objects belong to the context layer and which to the focus. The context layer is temporary and only shown during the gesture used to create and control it. This avoids overloading the screen with context information or allocating valuable screen real-estate to context information when it is not needed.
The view of the focus shown in image 3 on the Colour Plate shows a view that the user might see after having navigated for a while in the ZUI. This view does not contain any clues that would let the user know what map or chromosome is visible. In this situation the user can ask the ZUI to show the context layer. The context layer contains the initial view or context (image 1) and is drawn on top of the focus (image 3) giving image 6. The position of the focus is indicated on the context layer by a green rectangle. In image 6 this rectangle covers the text `10q' and tells the user that the focus is showing the chromosome 10. In image 7 the user has zoomed the context layer so that it shows the names of the genetic maps (the focus never changes during the use of the context layer). The green rectangle showing the position of the focus covers the name of the G'en'ethon genetic map. The focus is thus showing this map. Using the context layer the user has been able to position the focus in two different contexts.
When the context layer is visible the user can choose to concentrate on either the focus or the context by changing the relative transparency of these two views. The relative transparency can be continuously adjusted from a state where only the focus is visible to a state where only the context is visible. Image 8 is similar to image 7 except that the user is now contentrating on the context and has faded out the focus. The rectangle that shows the position of the focus is always visible and the user can see more clearly that the focus is currently showing the G'en'ethon genetic map.
Transparent overview layers [4] are a different type of display that differs from ours in that: (1) their layer is permanent while ours is a temporary orientation aid; (2) the transparency level of their layer cannot be changed by the user; (3) their layer always shows the top level view of the information space while ours can be used even if there is no top level view; and, (4) their layer can be used to move or modify the objects in the information space while our ZUI does not allow objects to be manipulated in this way.
The path taken by the user in the ZUI is a sequence of views of the information space. The first view is the initial (or top level) view (image 1 on the Colour Plate) and the view on the screen is the last current view (image 5). All the views (called the historical views) seen by users are stored in this sequence. Images 2 and 3 are historical views that the user has seen in going from the top level view to the last current view. The history layer is drawn over the top level view (giving image 4) and contains a view that can be varied by users from the last current view, via the all the historical views in order, to the initial view (and back again). The user can thus interactively `go back in time' and see the evolution of the current view in relation to the top level view. The comparison is done directly because transparent views are used so as to show the top level view and the historical view simultaneously. This comparison is also aided by the rectangles, drawn in two different colours, that show the sizes and positions, relative to the top level view, of the last current view and the historical view. The interactor used to control the history layer is described in section 5.
As with the context layer, the relative transparency levels of the history layer and the top level view can be adjusted so users can concentrate on the history layer or on the top level view. The two rectangles showing the position of the current view and the last current view are always drawn solid and are not affected by the level of transparency.
The current implementation of history overlays requires a top level view. This may not exist in systems where users can dezoom from a view of their own files to a view of, potentially and for example, the whole Internet. The scale of the top level view could also be so different from that where the user is currently working that users are unable to see changes to the positions of the current and historical views. We are currently investigating whether the system should choose a different view to replace the top level view and whether (and how) users can control this choice.
The techniques presented in the previous section help users to understand the information space from the top level view to the current position. However, users do not know what is in other parts of the information space and in particular what is to be found by further zooming. The user is also unable to use the hierarchy to navigate in the information space. Users looking at the details of an object are unable to rapidly dezoom to see the entire object and cannot easily move from a sub-object to another sub-object of the same type.
ZUIs are three dimensional spaces [7] and the scale is the vertical dimension. The main view in a ZUI shows a horizontal slice through the information space. We propose a second orthogonal view, called a hierarchy tree, that is a view of a flattened vertical slice through the information space. The names of those objects above the current position of the user in the space are shown. Objects also have types (an object can be a chromosome, a map, a sequence, etc). If the information space is highly regular, the entire hierarchy of types can be displayed in the hierarchy tree, otherwise just that part of the hierarchy centred on the user's current position. This hierarchy tree will thus show users the structure of the information space, what information is available, where that information is located and how to find it.
Figure 1: hierarchy tree
Figure 1 shows part of the main view of the information space plus the hierarchy tree. In this regular information space, the chromosomes are visible on the top level view of the information space. The chromosomes consist of arms, data and maps. This structure is displayed as soon as the ZUI starts. The user's current position in the structure is shown in magenta (or light gray); in this example the user is currently looking at the `Généthon' map on the chromosome 9. The structure indicates that if the user continues to zoom on the map, it will be possible to find the markers' sequences.
Excentric Labeling [5] offers a way of identifying objects on the screen. This technique labels, with `tool tips' in the main view, those objects located around the cursor. We propose a different, non intrusive way, of identifying the object currently under the cursor. As users move the cursor across the main view the hierarchy tree is updated to show the type and name of the object under the cursor. If the cursor leaves the window or is not on an object, the hierarchy view indicates the lowest level in the hierarchy to which all the objects in the main view belong. Our method of labelling remains similar to `tool tips' in that the user does not have to ask for the information to appear.
The gIBIS system [3] provides a global view of the displayed IBIS graph structure that is in some ways similar to our hierarchy trees. Their global view shows the subject of all the nodes in the network organised by their primary link. As users zoom or pan the local view of the network, the global view scrolls to show the users' their current position in the network. The global view can also be used to navigate in the network. Their global view does however differ from our hierarchy trees in that it shows all the nodes in the network. This means that at any moment only a small proportion of its contents are visible and a scroll bar must be used to navigate within the global view. ZUIs typically contain a very large number of objects and so a global view of all the objects in the information space would be so big as to be ineffective. As discussed above, our hierarchy trees are designed to make use of the structure present in many ZUIs and thus show the types of the objects in the information space and the names of only that object under the cursor and its ancestors in the hierarchy.
Ordering effects were taken into account. Half of the subjects answered their first 11 questions with the hierarchy trees and the other half of the subjects answered their first 11 questions without the hierarchy trees. Half of the subjects were given the first 11 questions in the list of 22 to do first while the other half of the subjects did the second half of the questions first. This lead to four equally sized groups of subjects.
For each subject we calculated the time taken to answer 11 questions without the hierarchy trees divided by the time taken to answer the other 11 questions with the hierarchy trees. A value greater than one from this calculation would mean that having the context aids was an advantage. The mean value was 1.58 with a standard deviation of 0.54. The high standard deviation was caused by the lack of familiarity of some of the subjects with ZUIs. For these people the training session was not long enough and they thus found the second set of questions easier. In general however the subjects were faster with the hierarchy trees and were positive in their comments regarding these aids: in fact those that started with the hierarchy trees were reluctant to continue the experiment without them. The other new techniques presented in this paper are currently being evaluated.
Pop-up and marking menus do not allow the operation to be controlled once it has been chosen from the menu (a scroll operation cannot be completed using just a standard menu) and do not allow users to supply the parameters required by the operation. An operation such as a font size change in a word processor often requires a dialog box to provide the new size. The users must select the operation using the menu and then concentrate on a second interactor. Once the new size has been entered the dialog box disappears and the users must reconcentrate on the main task.
Panning in a ZUI requires either a dedicated mouse button so that users can drag the image or two scroll bars. It is not possible to pan using a standard menu except with difficult to use commands such as `move a little to the left.' Zooming is also difficult to perform with a standard menu as users want to zoom until they reach the required scale. Standard menus only allow users to zoom by fixed steps and that only by repeated use of the menu.
Figure 2: the control menu in our ZUI
A control menu works somewhat like a marking menu. The novice user presses the mouse button and waits (0.3 seconds) until the menu appears under the cursor and then moves the cursor in the direction of the desired operation. The menu disappears and the operation starts as soon as the cursor has been moved the activation distance from the centre of the menu. (We have empirically chosen an activation distance of five times the radius of the circle in the centre of the menu.) The operation finishes when the user releases the mouse button. A user that has learnt the position of the desired operation does not have to wait to see the menu and can move the cursor immediately. The gesture is otherwise the same. This user has thus learnt to be an expert and is not distracted by the now unnecessary menu.
Figure 3: zooming with a control menu
During the zoom operation, the user can undo the current zoom by moving the mouse up or down a large distance. The user can then confirm the undo by releasing the mouse button or can undo the undo by moving the mouse back towards the centre of the display. It this case the user still has the mouse button pressed and can continue the zoom.
Figure 4: client server implementation
The interaction complexities that exist in ZUIs and especially in our ZUI with its extra context aids lead us to develop a new type of menu that combines the selection and control of operations. This new menu has been integrated into our ZUI but we believe that it will be useful in other visualization programs.
We hope that the combination of our new context aids and our new interaction methods will give users better control over a better interface.
This work was supported by the
European Union (contract BIO4-CT96-0346)
and by
the CNET
(contract 97 754 21).
8. REFERENCES
