Figure 11
This is one of the six original designs done in BadWindows. This design only shows topics and makes no effort to distinguish implicit from explicit information. As shown on the left-side scale, the items near the top are high priority while those at the bottom are low priority. The gradient background and use of color allows the high priority items to be emphasized with high contrast, as the low priority items fade into the background. Users can slide items up and down. The X-axis has no meaning. The boxes that contain more than one item are meant to convey a correlated topic. The header appears on the top while the items within the box are organized in a miniature version of the grid (higher is better, etc.).
In this earlier 1994 work, the newspaper appeared in one monitor while the user model was displayed on another monitor. When the section of a newspaper was selected, the "section" view of the model was shown. The user can slide any of the Topics up/down to raise/lower its importance. The buttons along the left side show available options (some are grayed-out):
In this earlier 1994 work (described in Figure 12), when an article was selected, this "article" model window would replace the "section" model window. In the upper-right, the top reasons why the article was selected appear. The thumbs on the left are for the user to explicitly provide quick feedback about the article with an option to attach comments. The matches between the article and the user model are shown in the main area. Users can provide feedback by sliding any item up/down. The Topics that appeared in the article, but not the model appear at the bottom. Users can choose to add these items to their model. The buttons along the left side show available options:
This is one of the abstract attempts to represent user modelling information. The vertical axis is preference, words nearer to the top of the screen are more preferred by the user. Each block represents some event that influenced the system's inference. The left column of blocks represent the positive influences, while the right column contains the negative influences. Notice how the word appears at the vertical point of difference between the two columns.
The colors represent the kind of influence (explicit, implicit, community, etc.). The fonts represent the categories of Topic, Type, Nature, and Source. Transparency shows confidence; the low confidence items fade into the background. This design is just too complicated.
As above, the fonts represent the categories of Topic, Type, Nature, and Source. Transparency shows confidence; the low confidence items fade into the background. In this design, the colors and shapes combine to represent the kind of influence (explicit, implicit, community, etc.). For example, in Figure 14 a green item may be an explicit influence but in this case a green triangle represents an explicit influence. The Y-axis is still preference. The positive influences push up from the bottom, the negative influences push down from above. The words that appear nearer to the top are more strongly preferred. Along with Figure 14, this design is overly complex.
The basic representation is exactly the same as the previous figure. The addition to the design is the angled text that appears behind each "influence shape". This text is a description of the influencing event. For instance, if a red square represents an implicit influence, then the text behind the red square describes when and how that influence occurred. Perhaps it was an observed instance of the user preferring a certain article. With this representation, users could find out precisely which data is being recorded about their actions and could also observe the impact of their actions on their user model.
To read the angled text, the user would move the viewer to a sideways view.
Although there is a great deal of information in this compact representation, the design is unbearably complicated.
This is another variation on Figure 15. In this case, a plane is drawn in 3D space to connect correlated items. Since no added information is shown about how the correlated items compare to each other, this is not a particularly useful design.
The main window of the "slices" design is shown here. The distinctions between this design and the currently implemented version of PeerGlass are the use of a sidebar and the display of only one totem pole. The sidebar is a very useful browsing device. There are four subsections to each section, and at most six articles in each subsection. By clicking on the sidebar, the user can select which subsection is displayed.
(Please disregard the blank areas below, they do not affect the design.)
This is a view of the model planes in the "slices" design. The only difference between these model planes and those in the currently implemented Rolodex design is their orientation. A problem with the slices design is that only a few planes can be displayed legibly in the stacked layout. The user can click the "Examine Profile" button to enter a mode where the arrow keys select a plane from the stack. Once the user has selected a plane, pressing the Enter key will launch an animation that brings the selected plane to the front and pushes all others downward. From this point it is assumed that the user can zoom into the current plane for closer viewing. The modal nature of the slices interface is a major drawback.
A close-up view of a model plane from the "slices" design (see the caption of Figure 19)
When an article is currently selected, the one totem pole from the "slices" design displays the strongest matches between the article and the user's model. If a subsection or section is currently selected, then the totem pole displays a summary of the totem poles for the set of articles in the subsection or section, respectively.
Figure 12
Figure 13
Figure 14
Figure 15
Figure 16
Figure 17
Figure 18
Figure 19
Figure 20
Figure 21