To increase bandwidth, we explored the use of subsymbolic user behavior, such as mouse movement, scrolling, or mouse clicks, for user modeling in AH. In order to make subsymbolic user behavior available for user modeling we had to overcome several problems, including how to collect and analyze this kind of data. For collecting subsymbolic user behavior, we have developed a novel method that minimizes interference with normal user behavior and environment. Our method can be used to collect subsymbolic user behavior in virtually any web-based AH. We analyze the collected data using Self-Organizing Maps (SOM) for exploratory data analysis. Using this approach, we visualize and analyze the structure of features extracted from the collected data. We tested our method in a field study on an existing web-based AH, the Adaptive Collaborative UNIX Tutorial (ACUT). Using our method, we constructed features for user modeling that capture the underlying structure of ACUT's user population. The case study demonstrates how the developed approach forms a solid basis for user modeling. Our results support the hypothesis that subsymbolic user behavior constitutes a promising source of data for user modeling in AH. Because our method does not rely on any application specific assumptions, it is more generalizable than established methods.

Refreshments will be served.