5.1 Basic Issues of Active Information Systems

Implementing active information systems is quite different from implementing passive information systems that support browsing and querying only. In passive information systems, the process of information seeking is explicitly initiated by users, and the needs for information are either articulated as retrieval queries or externalized through a series of browsing actions. In active information systems, the system must determine the information needs of users and when and how to present the retrieved information.

5.1.1 Contextualization: What to Deliver?

For users who are engaged in a task, most of the time they are not very interested in information that bears no relationship to their current task. They need only information that helps them accomplish their task. Because different users have different knowledge backgrounds, their needs for information are also different. For most active information systems, the critical challenge is the contextualization of new information to the task acted upon and the user acting.

Active information systems that just throw a piece of de-contextualized information, such as Microsoft's ``Tip of the Day'', are of little use to most users. This type of system could be viewed as a reverse help system that exploits the communication paradigm of ``Answer First, Then Questions'' in contrast to the traditional ``Question-Answer'' paradigm of most help systems [Owen, 1986]. Despite the possibility for interesting serendipitous encounters of information [Roberts, 1989], most users find this feature more annoying than helpful. The random presentation of information also makes it difficult to understand when or how the information should be used due to the lack of the problem context.

Sections 5.2 through 5.6 explain in detail how to achieve this contextualization in active information systems, which is the focus of this dissertation research.

5.1.2 Feedforward or Feedback: When to Deliver?

Depending on the temporal order between the time when the information is delivered and the time when the user action for which the information is delivered takes place, active information systems can provide feedforward or feedback to users.

For each action, there is a period of time called action-present, in which users have decided what to do but have not yet executed the needed operations to change the situation [Schön, 1983]. Information delivered in this period of time is feedforward [Simon, 1996] information because it can make users change the course of action or assist users in accomplishing the action (Figure 5.1).

Figure 5.1: Feedforward information delivery

$\parbox{.8\linewidth}{\small{Information delivered during the period... ...esent is feedforward information that can affect the execution of the action.}}$

For example, the Autocompletion of Internet Explorer provides feedforward (Figure 5.2) to users who want to visit a website by saving some keystrokes, but more importantly, by relieving users from remembering exact addresses of websites.

Figure 5.2: Autocompletion in Internet Explorer

$\parbox{.8\linewidth}{\small{When a user types http://www.cs into th... ...string are shown in the pop-up menu, and the user can choose one to revisit.}}$

Feedback information is delivered when the action for which the information is delivered has been finished (Figure 5.3). Feedback can create a situational backtalk of the action by pointing out a potential breakdown the user has not known or noticed, or can augment the situational backtalk to help users reflect better on the action just completed [Nakakoji et al., 1998]. Feedback can serve two roles. First, it creates a learning opportunity for users to improve work performance. For example, the ACTIVIST system [Fischer et al., 1985] teaches users the corresponding key shortcut to replace a series of complex keystrokes used in their previous action in a text editor. Second, if the previous problematic action can be undone or modified, it helps users reach a better solution, such as the on-the-fly spell-checking mechanism in many word-processing systems.

Figure 5.3: Feedback information delivery

$\parbox{.8\linewidth}{\small{\small{Information delivered after the ... ...en finished is feedback information that help users reflect upon the action.}}}$

Feedback is retrospective because it gives users a chance to change a problematical or suboptimal solution; feedforward is prophylactic because it prevents a problematical or suboptimal solution. To provide feedback, systems have to compare users' solutions with ideal solutions to find out what went wrong; to provide feedforward, systems have to predict what is needed by the user in the near future based on what has been done so far.

5.1.3 Interruptive or Noninterruptive: How to Deliver?

Because information delivered by active information systems is unsolicited, it has the risk of interrupting the workflow of users whose primary goal is not the process of the delivered information. When delivered information distracts users, it becomes intrusive. The intrusiveness of a system is the degree of users' perception of being interrupted from their current focus. Not all intrusive information is bad. Information that prevents a user from making a mistake that may cause all subsequent work to be void needs to be timely attended so the user can avoid the cost of revising a whole chain of action.

Information can be delivered interruptively or noninterruptively. An interruptive delivery requires the immediate reaction of users: if users do not attend to the delivered information, they cannot continue their current work. A noninterruptive delivery just presents information with no reaction from users required. It is up to the user whether to pay attention to the delivered information. Although noninterruptive deliveries present less disruption to the workflow of users, they may go unnoticed and provide insufficient help. Noninterruptive delivery can have various degrees of intrusiveness, depending on how the delivered information is presented, for instance, the distance between the window displaying the information and the focal window of users. On one end, if the window does not exist or is hidden from the current working space and gets opened or displayed only when the users become interested, the intrusiveness does not exist. LispCritic [Fischer and Mastaglio, 1989] is an example of this type of system. On the other end, if the information window is placed right in the middle of user's current focus, the intrusiveness is close to interruptive delivery.

Active information systems need to achieve the right balance between the cost of intrusive interruptions and the loss of context-sensitivity of deferred alerts [Horvitz et al., 1999] by carefully considering when and how to deliver the information so that it can be utilized best by users. Depending on the importance of the information, systems can explore a variety of intervention modes to decide when and how to interrupt the user [Sumner, 1995].