After the preprocessing step, behavior representation is performed
for the derived access sessions of elder self-care in Com-
Care. Frequency-based representations use a frequency vector in
which each element corresponding to a given self-care function
is computed by counting the uses for that function and dividing
it by the total number of uses in a session. Clustering studies of
Web usage mining often apply frequency-based representations
because of their easiness for representing user sessions and for calculating
the distances between sessions. Here, sequence information
for elder self-care behavior is captured by a sequence
matrix, i.e., a matrix representation constructed based on a transition
matrix of a Markov chain similar to that developed by