A state of a transition matrix of a Markov chain can be defined
as either a sub-function or as a main-function (a category of subfunction)
that can be used by the elder for self-care. In Park
et al., user sessions are represented using categories of general
topics, not pages themselves, and each Markov chain represents
the behavior of a particular subgroup (Park et al., 2008). Fu et al.
showed that clustering can also be performed in the sub-function
category by using a distance-based hierarchical clustering
algorithm to cluster user sessions. Although each ComCare subfunction
can be considered a state, considering a main-function
as one state is preferable for two reasons. First, the self-care behavior
of the ComCare participant within the same category of
sub-functions may not provide sufficient information to identify
the elder clusters. Secondly, since the performance of the clustering
algorithm decreases as the number of sub-functions grows
and ComCare contains 71 sub-functions, the clustering problem
may be unmanageable if all sub-functions are considered states.
In this discussion, therefore, the term ‘‘sub-function’’ actually
refers to a main function.