Premise: to have an idea of the experiment that I am currently analyzing, refer to this technical report.
These days I have been working on creating a meaningful visualization for the eye-gaze sequences that we collected in my current experiment on collaborative problem solving at distance. As explained in my previous post, the initial step to present the huge amount of data that we collected in a meaningful way consisted in constructing a time-series of the time intervals of editing of the utterances produced and the icon movements across the work area.
This time-series per se is interesting to compare different experimental conditions. However, does not explain what the peers were doing during the editing of the messages. To introduce this information, I used the eye-tracker data dividing the work area in top-level polygons: the map area; and the chat window, which was subdivided in history pane and composition pane. I mapped these three polygons to the basic activities of: identifying (map), reading (history), and writing (composition box).
Then I aggregated the eye-gaze data in sequences of permanence into one of these areas. Then, given the start time and the end time of each sequence I could complete the time-series superimposing a timeline visualization of these three basic activities.
An excerpt of the resulting image is shown below. Brown color represent the identifying activity, red is reading and blue writing. For each utterance I report the activities of both peer using the upper line for participant 1 and the lower line for participant 2. What emerges at a first glance is that during the message composition there is an high probability that the eye will jump on the map and on the history whenever the message utilizes geographical references.
In other words, whenever the message composition requires a geographical encoding of information which are taken from the context (previous messages and / or shared landmarks), the eye checks these information along the editing of the message that should transmit this information. In the sequence of the 4 messages reported below, the last one shows exactly this phenomena: while the first participant is busy exploring the map the other is composing a proposition in which s/he acknowledge the choice of the two parking introduced by the first participant in the first utterance of the sequence, and s/he also introduce an extra arrangement of the scenes based on this selection. At the beginning of the message composition, the eye quickly glance at the history, looking for the confirmation of the parking labels (185 and 180), then quickly at the map looking for the path between the parking lots and the stages. Finally s/he spends the remaining time in the message composition window.
