User Interaction

User interaction with Calvin occurs through one of three different interfaces. The ``User Interface'' is designed to be used by an end user in day to day interaction with Calvin. The user interface used in our research is shown in figure 1. Through this interface, Calvin suggests resources to the user. The user can also see a list of the resources Calvin has accumulated from the current browsing session. Users can set up customizable filters to prevent indexing particular kinds of documents (such as documents containing the user's name, or advertisements from web browsers). Different data analysis components may also have various user-controlled parameters, or special diagnostic displays to show the user. The user interface is designed such that different data analysis component can provide different displays to the user without requiring changes to the user interface itself.

  

Figure 3: Task Generator Interface

When performing experiments to study the behavior of a data analysis technique, one may not want the user to receive so much information, or have such control over the system. In addition, it may be necessary to request additional user feedback. To facilitate collecting data for experiments, Calvin includes a ``Task Generator'' which provides a different user interface to the Calvin system. This interface presents the user with a series of tasks which the experimenter determines beforehand. After the user spends some specified amount of time on the task, a new window is displayed to ask the user experimenter-defined questions. Figure 3 shows a sample survey developed for an experiment on user browsing behavior.

During the experiment, the system keeps records not only of responses, but also of the task context in which those responses occurred. Information about every document accessed is gathered and stored, including the document location, full text of the document (if available) and document title, and a time stamp when the document was accessed. This information, along with the answers to the surveys are cross indexed according to an internally assigned id number and unique document id.