Lincoln Ritter's Idea
During a recorded tutoring session, the student generates strings of
events representing his/her path through the problem/solution
space. Further, a "correct" solution can be expressed as a
particular subset of all the possible strings over the alphabet of
recordable events. To effectively extract useful information from
student generated event logs, two things must be known. First,
some knowledge of a correct path from problem to goal must be
developed. Regardless of the approach taken to analyze student
data, an idea of when a student has veered off a productive path must
be had if effective interactive tutoring is to take place.
Second, some way of comparing paths to, at minimum, determine some
notion of distance between correct paths (paths that will more likely
lead to a correct solution) and less correct paths.
To this end, I propose that expert users be elisted to solve problems
so that "correct" paths through the space can be recorded. Then
using existing student data, we can perform gapped alignment algorithms
to compare student paths with expert paths. presumably, students
who wandered insearch of a solution will generate event strings that
are more dissimilar to the expert paths than students who who navigate
to the solution succinctly. Insights gained from these
experiments could then be used directly, or as a starting point for
other data mining methods.