Figures 1 - 10 (watch out for bad colormaps)

Figure 1

The Sources that strongly match the current article and the user's model appear on this side of the totem pole. In this case, the article about Psion appeared as a featured article in Byte magazine. About five Sources can be neatly listed here.

Figure 2

The Psion article covers the Topics of: palmtop, PDA, technology, and computers. There are other Topics that appear in the article but only those that strongly match the user's model appear here. About eight can be legibly listed.

Figure 3

The Types of news that appear in the Psion article and match the user's model are: Press Release and Review. The angles for this side of the totem pole are the hardest to read. The difficulty is that changing the angles may cut-off more of each word. Still, some improvements are necessary.

Figure 4

Adjectives that describe the Nature of an article appear on this side of the totem pole. The Psion article is Promotional and Consumer-oriented. Both of these descriptions match the user's model. At most, six adjectives can fit well on this side.

Figure 5

This is what the user sees when PeerGlass is launched. The section choices appear down the left side. Which sections are listed is dependent on the user's model and the available news. Currently, the "Tech" section is being displayed. The six totem poles correspond to the six articles shown. There is text on the bottom of the screen explaining the purpose of the totem poles and instructing the user of the available options. The user has four options now:

1. Change sections by clicking on a section button.
2. Display the full text of an article by clicking on the article or its totem pole.
3. Rotate or manipulate a totem pole with the mouse or arrow keys. This will provide insight about why the article was selected.
4. Click the "Examine Profile" button to see the user model for this section.

Figure 6

The Rolodex of model planes becomes visible after the "Examine Profile" button is clicked. Above the Rolodex, there is a paragraph explaining the contents of the current model plane. In this example, the current plane is the Implicit observations plane. Below the Rolodex, is text that explains the use of the Rolodex. The totem poles remain visible so that the user is able to compare their content to the model planes. The options available to the user are:

1. Spin the Rolodex with the dial.
2. See the next/previous model plane using the arrow keys.
3*. Change sections by clicking on a section button.
4*. Display the full text of an article by clicking on its totem pole.
5*. Rotate or manipulate a totem pole with the mouse or arrow keys.

Figure 7

The Explicit interests plane lists the interests that the user has told the system to include. The categories of Type, Topic, and Nature appear on this plane. Since the levels of preference and confidence do not vary with explicit information, the preferences appear in alphabetical order. In response to the user evaluation, there will be a scale of preference in future versions. Also, the design will have to be altered to fit more than a dozen items in each list. Currently, no interface is implemented for the user to explicitly select interests but there are a variety of possible interfaces discussed later in this document.

Figure 8

The Explicit expert sources plane shows the sources that the user has told the system to include. The vertical scale along each column represents the priority assigned to articles by the expert source. Although the priority scale for most newspapers will be the same, it may vary for magazines and less frequent publications. Without the priority scale, the system cannot determine the importance of an article. If the user does not declare explicit expert sources, it is impossible for the system to incorporate headline news that involves fresh stories (not likely to match the existing set of explicit or implicit interests). Declaring favorite sources also helps to prevent redundancy in article selection.

Figure 9

The Implicit interests plane shows the results of inferences the system has made by watching the user's behavior. If the user seems to like a certain Topic/Type/Nature/Source of news, then that item is presented near the top of the plane. As the system confirms these inferences by making more observations, the item appears larger. The user study has shown that this plane is very confusing for many users. New, clearer designs are presented later in this document. The goal of these designs is to present the user with the high confidence inferences and avoid distracting the user with low confidence, non-influential items. Another upcoming change is the removal of the "off-sides" view, currently showing the spiral.

Figure 10

The Implicit communities plane shows communities that are formed by readers who share common interests. Each community is represented by a triplet: three items that represents the community's most salient interests (Topic, Type, Nature, or Source). The triplets that appear closest to the word "you" in the center are very similar to the user's interests. The triplets that are small and far away, are much less similar to the user. The trouble with this design is that it is hard to read the triplets that are close to "you". If a few more triplets were added near the center, the plane would be illegible. New designs are presented later in this document.