the qualitative analysis reveals that there are correlations
among the themes. For example, “heavy study
load” can cause “lack of social engagement” and “sleep
problems”. Also, “negative emotion” may be associated
with several other themes. The Naıve Bayes classifier is built
on top of the label independence assumption, which is a
simplification of the real world problem. The classifier is
designed to be a multi-label classifier in order to reconcile
this effect. If a tweet expresses correlation between “heavy
study load” and “sleep problems”, then it can be categorized
into both categories. After all, any mathematical and
statistical models are simplification of real world problems
to a certain extent. The comparison experiment with M3L
shows that this advanced model that accounts for label correlation
does not perform as well as the simple Naıve Bayes
model. Future work could specifically address the correlations
among these student problems.