Each elder access session is eventually transformed into an mdimensional
vector of interests of sub-functions, i.e., s = {int1,
int2, ..., intm}, where m is the number of sub-functions used in all
user access sessions. However, if the number of dimensions m exceeds
a reasonable size, it not only consumes substantial processing
time during clustering sessions, but also limits the real-world
applicability of the system. Dimensions are reduced by using a frequency
threshold fmin as a constraint to filter out sub-functions that
are accessed less than fmin times in all access sessions. Our research
showed that 60% of sub-functions appearing in the access sessions