Of relevance to this work is the flexibility to communicate information collected by agents in the appropriate manner. An example of this is Apple Data Detectors, a series of tools that scan selected text and offer potential actions based on the structure of the text (see [57]). For example, a user can open or bookmark URLs, send email, read newsgroups, or retrieve files from FTP sites. Other such agents can be developed by Agentsheets, a spreadsheet-like visual programming application for creating information-triggered agents [65]. Examples of agents created by Agentsheets range from educational simulations of bridge building and electricity to information processing systems that interpolate temperature over a large region based on a few points.
The contribution of the work in this thesis is not in the creation and understanding of agent technologies, but in the development of awareness techniques for the type of information collected by these agents. As information sources found on the World Wide Web and similar repositories become broader in scope and more frequently updated, the need for appropriate communication techniques will become even more pronounced.
Monitoring is probably the more widely addressed and widely understood of the two, with numerous examples of monitoring systems available. Jonathan Cohen provides a support structure for monitoring based on audio alerts [16]. More recently, work by Lyn Bartram has explored the perceptual properties of motion as it proves useful in communicating information [5]. In both examples, the communication technique is designed to draw attention away from the current task to some other task, useful in a monitoring situation.
Awareness typically involves a desire to acquire or maintain knowledge about changing information without distracting the user from other tasks. In studying the communication properties of animated effects, Paul Maglio is careful to focus only on information gained on the periphery, while at the same time measuring the distraction peripheral displays may cause in performing a primary task of editing a document [45]. However, studies like these may create an unrealistic testing environment given that the distinction between monitoring and awareness is not firm. For example, radios, cuckoo clocks, and television could be viewed as either monitoring devices or awareness devices. Cadiz's Awareness Monitor system semantically and functionally emphasizes the blending of these concepts, using a series of effects (including tickers) to address both monitoring and awareness issues [11]. Monitoring and awareness tasks are often closely related. For example, individuals in general may want to be aware of a stock quote, but during critical times they may want to monitor it. Similarly, when keeping track of the score of a ball game, in general it may be desirable to be aware that one team is far ahead of the other or that it is early in the game, but if the score remains close near the end of the game a viewer might want to focus attention on the game. While the main focus of this thesis is on awareness, at times the tasks I chose to examine cross into what many would consider to be monitoring.
Recent work has focused on non-desktop methods for maintaining awareness. Hiroshi Ishii's Tangible Media Group at MIT coupled information with everyday physical objects to communicate information at the periphery using light, sound, air flow, and water movement [35]. Scott Hudson at CMU adopted a similar approach, but he focused on techniques that do not rely on the projection of light in order to lessen the possibility for distraction [30]. Background audio also has been used to assist in monitoring and awareness in Mynatt's Audio Aura and Cohen's auditory display system [56, 16]. Though many of the techniques described in this thesis are applicable to non-desktop systems, to simplify testing and increase the potential user population this thesis focuses on displays in the periphery of the desktop.
Desktop communication often focuses on graphical displays of information. A well-designed graphical display can communicate information using a small amount of space. Graphical communication mechanisms have been making their way onto computer desktops as the speed and processing capabilities have made them viable. The rules and guidelines for the use of graphics established and explained by Edward Tufte [77] suggest that communication using graphics can be advantageous because more information can be communicated in a smaller space than with text alone, and because graphical images can often be processed with less cognitive effort than can textual ones.
In cases where the information to be communicated is graphical, graphical images should be used to convey the information. This would allow the information to be conveyed with minimal changes to its form, minimizing distortion of its meaning. For example, a Web site that consists largely of images might be best represented by showing parts or all of these images when they change. This technique was used in Helfman's passive Web browser that used pictures from a proxy that had been loaded by other users [31, 32]. A similar technique was used in Andruid Kerne's Collage Machine system [39] and Brown and McCrickard's CWIC system [9]. However, the drawback to these systems is that images from Web sites are typically large and require a fair amount of precious screen space. These systems overcome that either by requiring the user to read mail [31], visit a Web page [9, 39], allow a screen saver to run [9, 63], or view the screen wallpaper [9] to view the images. Still, this does not constantly raise awareness while allowing the user to perform other tasks using the same screen real estate.
One way to raise awareness in a small space is with icons. An icon is a small picture used on buttons and other widgets to communicate information about functionality or contents. For example, the``mailbox flag down'' icon on many email monitoring programs programs (for example, the Unix xbiff program [25]) indicates that there are no unread messages, while the ``flag up'' icon indicates that there are messages. Of particular relevance to this work are animated icons, small images that change to represent actions [4]. For example, a drawing action may be difficult to represent in a graphics palette with a static icon, but an animated icon can show each of the steps in the action. Empirical evaluations have shown that the use of animated icons to represent complex word processing tasks results in improved performance [6]. However, much of the information that needs to be communicated is textual, and it is acknowledged to be difficult to generate icons for arbitrary textual information [4].
As such, the focus of this work turned to techniques that would support textual information. Of course, the earliest information monitoring and awareness tools, which ran on text-only displays, had to use text to communicate information. The biff email tool interrupts the user's current activity by printing the header and first few lines of a new email message in a terminal window [78]. In a similar manner, the wall tool displays information such as machine shutdown times on the terminals of all currently logged-in users [80]. In windowing systems, tools such as plan demand immediate attention by popping up windows containing textual messages about events on one's personal calendar [18]. These tools fall more in the category of monitoring tools, with the goal being to interrupt the current task of the user with one of higher immediate importance.
While this is an important category, it is beneficial to consider less intrusive methods. One such method is smooth, constant animation. It is widely acknowledged that this type of animation can become less intrusive over time, and the use of animation has interesting potential in maintaining awareness. The next section looks at ways in which animation has been used as a communication mechanism.
Studies have shown that animation can be useful in certain decision-making situations [27]. In these studies, participants completed a home selection problem and a physics problem, and they were evaluated based on completion time, accuracy, ease of use, and enjoyment. Participants performed better with smooth, realistic, interactive animations compared to abstract representations and abrupt changes. In these and similar situations, the animation often has a significant visual (often physical) component: either it deals with a physical object, or it has a commonly-associated visual representation. It is unclear from these results whether the use of animation will be effective in broader, more abstract problem domains such as information communication.
One such abstract problem domain is algorithm learning and understanding. While Stasko's path-transition paradigm for algorithm animation was developed with the expectation that smooth animation rather than sudden changes would help to communicate the information in a more understandable manner [72], initial empirical research on its benefits showed disappointing results [74, 10]. However, other work has been more promising, particularly in hands-on situations where users have significant control over the algorithm animation. One study showed that the use of animation was beneficial when students were permitted to enter their own data as input [42]. Another study showed that algorithm animations proved particularly useful in interactive learning situations [38]. However, a high degree of interactivity cannot be expected in the secondary monitoring and awareness situations explored in this thesis.
The communication abilities of animation in other learning situations have been explored as well. A study compared the use of animated demonstrations, written text, and a narrated animation for teaching users how to operate a particular graphical interface [61, 60]. Although participants completed the initial training task more quickly when using animation, they completed the testing task more slowly. Furthermore, after a week, participants who used the animation performed more poorly on the task. This work showed that animation can be effective in the short term for learning, but often the learned information is not retained in the long term, perhaps because participants merely mimicked the behavior they saw in the animation rather than attempting to learn the proper behavior. It is worth noting, however, that the learning tasks in this situation are quite different from the awareness tasks explored in this thesis. It is expected that in awareness tasks, users will use the animation not to learn new information, but merely to recognize that a change has occurred. They will use more suitable interfaces to learn the information if desired.
While the prior work described in this section suggests that animation may be useful in other domains, it is difficult to make direct connections between the results described here and those expected for peripheral awareness tasks. The animated displays described in this section are typically either the primary focus of the user or are designed to draw the attention of the user to the animated information. Furthermore, one common conclusion is that animations are more effective when they are highly interactive, an undesirable trait for secondary, peripheral displays. Most importantly, the tasks that the users are trying to accomplish (learning and decision making) typically are more cognitively demanding than the monitoring and awareness tasks considered in this thesis. Little work has been done in understanding the usefulness of continuous, long-term animation for secondary awareness tasks.
Recent work has seen attempts to include animation in awareness devices, but many of these devices are dismissed as being too distracting, and little testing has been done to understand their potential utility. Constant, cyclic animations such as those studied in this thesis can be seen in stock quote displays, advertisements, and sports scores used by many programs and on many Web sites. For example, Yahoo released a stock ticker in 1996 that has experienced several iterations and is still available (see [81]). Pointcast is perhaps most widely remembered as a user of animation, employing horizontal scrolling effects and other effects in their screensaver and standalone application [63]. Numerous Web sites, including the Golf Channel (www.thegolfchannel.com), the Atlanta Visitor's Bureau (www.avcb.com), and the Georgia Tech College of Computing (www.cc.gatech.edu) at one time used tickering effects to communicate information, but each has now abandoned the effect, no doubt after complaints from users. Rather than abandoning the use of animation entirely, this thesis attempts to understand how animation can be used effectively and when potential users would be more likely to utilize it.
Several widget sets have been developed to allow the programmer to include animation in the interface. The Artkit toolkit allows programmers to create transition objects that describe how an object will move [34]. A reference to the transition object is then added to a graphical object to create the animation, and Disney-like effects can be associated with the animated effect. Another toolkit, Amulet, was extended to include support for animation [55]. Animation can be attached to any value of any object, including position, color, or visibility. These and similar toolkits provide a great deal of power, yet they often can require significant effort by the programmer to achieve a desired result.
The widget set developed as part of this work does not require the user to specify the details of the animation. Instead, in only a few lines of code, a programmer can specify variables, define animation behavior, and start a cyclic, repetitive animation that automatically updates when variables are changed by the program. This same behavior would be much more difficult with most other toolkits. At the same time, the animated widgets are structured such that new animated widgets can be added with significant code reuse.
Some of the earliest evaluations of animated displays examined the perceptibility and readability of rapid serial visual presentations (RSVPs) of letters, strings, and words. Foster studied the readability of sentences that were presented a word at a time in a single visual location using motion picture film [23]. He found that graduate student participants could correctly identify about four out of six words in a sentence when presented at 62.5 milliseconds per word. Juola extended this word to also consider comprehension of information on computer screens presented as RSVPs and in multi-line paragraph format [36]. He found that comprehension was comparable in the two presentation modes.
The first studies of smooth animated effects were performed in the early 1980s. Duchnicky and Kolers performed a series of experiments examining the readability of text scrolled on visual display terminals as a function of window size [19]. They found that larger displays typically led to faster performance on reading tasks. A study led by Michael Granaas found that larger jumps (four to ten characters) led to better comprehension than smaller jumps (one to two characters) [28]. Kang and Muter compared a tickering effect to a non-animated ``blast'' effect where the information was changed without a gradual animation [37]. They found no difference in comprehension for a display reading task.
The previously mentioned evaluations all examined the reading of animated displays as the sole task of the participant. Only recently and concurrently with the work in this thesis has the usefulness of animation in maintaining awareness and in monitoring events been explored. Paul Maglio performed a series of experiments to examine the tradeoffs in displaying peripheral information [45]. Participants performed a series of primary tasks where they were asked to edit a document. Simultaneously, a continually scrolling, start-and-stop scrolling, or fading display would show information. He concluded that continually scrolling displays are more distracting than displays that start and stop, but information in both is remembered equally well. Scrolling direction does not seem to affect performance, and additional cues that are auditory have a more negative impact than additional visual cues.
While these results are interesting, they provide a narrow view of the usefulness and impact that peripheral animated displays can create. The work described in this thesis examines effects on a browsing task, a cognitively less demanding primary task that better matches the expected usage pattern noted in usage studies of animated displays (see [46, 22, 62, 52] and Section 5.3). These studies showed that many users would not use animated displays at all times, but rather only during selected low-effort primary activities. Also, the work in this thesis considers not only awareness activities but also monitoring activities that require participants to perform a simple task when they notice certain changes in information. This combination of activities again seems to better match the different types of activities that users would perform in the real world. Even though the Maglio work includes visual and auditory cues that are designed to draw attention to changes, this work does not specifically test whether participants notice these changes, only whether they are distracted by the additional cues.
Several early observational studies that I performed contributed to the direction of this thesis. The Irwin study examined how a set of communication techniques (audio, graphical encodings, animation) impacted awareness for four users of constantly changing resources such as email, Usenet news, weather, and certain Web sites [46]. The development of the tool and the study conducted to examine its use is described in Section 3.1. The tkscore study focused on animated and non-animated displays of ball scores during the single-elimination NCAA basketball tournament [52]. The reactions of over twenty users to various aspects of the different displays is described in Section 4.1.
Other similar work at the Georgia Tech GVU Center that is currently under way is an observational study by Alex Zhao of his What's Happening (WH) system. Much like the evaluation of the Irwin system, the WH evaluation examines how a set of communication techniques can help users maintain awareness of information, in this case information about local events that could help build a sense of community. Initial results of the study can be found in [82].
Numerous observational studies conducted elsewhere highlight guidelines for obtaining useful results. Notable among these is work such as the Bellcore VideoWindow study [21]. The VideoWindow provided an audio and a video link between two sites, then used cameras to observe how users interacted when using the VideoWindow as compared to face-to-face interactions. The large amount of data collected allowed the experimenters to make interesting conclusions about the reluctance of participants to use this new technology. By focusing on a limited set of related communication techniques, the authors were able to understand utility and establish guidelines for the techniques.
While the studies described previously in this section have examined a variety of communication mechanisms, fewer studies have focused on a single technique as in the case of this thesis, constant and cyclic animation. Perhaps the work most similar to the observational study described in this thesis came from the Elvin project [22, 62]. This project included a system called Tickertape that collects information from Web sites and newsgroups, then allows users to send and read messages that are displayed using a single line scrolling message window. An observational study looked at twenty Tickertape users who were interviewed in semi-structured sessions that often digressed into conversations. The result was a categorization of users based on the tasks that Tickertape helped to support: work, social activities, leisure, or news watching. The existence of categories were supported by sample dialogues and quotes from users. It was not clear if there were specific characteristics of users that made them more likely to fall into a certain category.
One problem with many of these studies is that they try to evaluate too many different techniques. For example, our experiences in evaluating Irwin showed that people found the tool informative and interesting, but it was unclear if this was because of the graphical encodings, the audio cues, the textual information, or a combination of these factors. Even the Tickertape study, which involved a system that used a single communication technique, was more of an exploratory study examining potential uses of an awareness system rather than an evaluation of the effectiveness of an animated display for different users. The goal of this thesis is to explore the effectiveness of a single class of techniques and to understand the different ways that potential users might react to animated techniques. In particular, the observational study categorizes user behaviors based on shared characteristics of users with similar preferences. See Chapter 7 for a full report of the study.