Analysis

Our systems easily predicted the
true labels of sentences which had either positive
connotation words like “great”, “pleasant”, “nice”,
“good”, etc or negative connotation words like
“not”, “slow”, “unable”, “horrible”, etc and classi-
fied them into comment and complaint classes respectively.
The negative connotation words also
appeared in the feedback sentences of bug class.
But owing to the larger amount of training data in
the complaint class as compared to the bug class,
the negative connotation words appeared signifi-
cantly in the complaint class. As a result, our
systems had difficulty in predicting the true labels
for the feedback sentences associated with the bug
class. Our systems were unable to detect some
tags due to the class imbalance problem in the
training as well as test data. The scores of our systems
could have been much better, provided that
we should have more labeled training data. The
system performance can be improved by the language
specific pre-trained word embeddings.