Stephan Weibelzahl

To navigate through web-sites (e.g., presentations, online-shops, or learning courses) the user requires a mental model of the site's structure. Many adaptive systems are targeted at minimizing the users' cognitive efforts during interaction. Thus, adaptive web-sites should either support the user in building an adequate mental model or minimize the model's size. In a series of experiments we were able to show that the interaction with an adaptive system (either an electronic catalog (N=25+17) or a learning course (N=72)) requires less complex mental models than the corresponding non-adaptive version. We observed individual behavioral complexity by modeling the human-computer interaction as a state-transition network: the system changes its current state (e.g., presents a new page) if the user initiates an action (i.e., a defined transition such as visiting a hyperlink or pressing a button). The mental model of users who navigate in circles (i.e., they return to a previous state and did not find the straight way to the goal state) is obviously incomplete or incorrect. The mental model of users who applied less transitions to reach the same goal-state is less complex and thus more easy to achieve. Individual state-transition networks allow for a simple computation of the number of interaction-circles and network-density as measures of complexity. Participants who were supported by an adaptive system had significantly less complex mental models than those who solved the same task with a non-adaptive version. We argue that behavioral complexity is an interesting measure for the evaluation of adaptive systems where other measures (e.g., duration or interaction-steps) frequently failed.