Page 144 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 5 Depression discovery in cancer communities using deep learning 133

as applying a balloting system based on the majority rule or stack-
ing different algorithms. The results outperform the results of in-
dividual data sets, thus proving the efficacy and usability of the
combined scheme in the presence of parallel data sets.
Áuthors in Ref. [50] utilize features from both a particular post
and its paternal post to perform SA on debate texts. They use ML
classifiers to catalog standpoint. Standpoint can be thought of as a
person's view toward an entity, idea, or situation. They use SVMs,
NB, and a rule-based classifier. Even after extensive experimenta-
tion, it was observed that the unigram baseline still outperformed
all the different combinations of dependency features. They
concluded that better features need to be determined.

2.3Hybrid approaches
Table 5.4 presents an overview of hybrid approaches for SA.
Authors in Refs. [51] propose an approach to SA in outsized tex-
tual databanks that utilizes discourse information and addresses
the problem of polarity shifts of terms depending on their usage
context. Most SA approaches have no way to deal with ambiguous
terms and ignore them, although a word may have more than one
different connotation and different polarities that change accord-
ing to the context [43]. To address this shortcoming, authors in
Ref. [51] use NB method to identify ambiguous words in text.
They develop a contextualized vocabulary enlisting the polarity
of the ambiguous words, along with adjacent context words.

Table 5.4 Literature Summary of Hybrid approaches for SA.

Lexicon and Level of
Authors algorithms used Data set/Source Polarity Language granularity
Weichselbraun General Inquirer Lexicon Reviews from amazon.com, Pos/neg English Document
et al. [51] and NB classifier Tripadvisor.com and Pang and level
Lee Movie Review Dataset
Agarwal et al. Dictionary of Affect in Tweets Pos/neg English Sentence
[52] Language and WordNet Pos/neg level
Lexicons, SVM classifier and
neutral
Fiadhi et al. Ding Lexicon and J48, Twitter Pos/neg English Sentence
[53] JRIP, K*, NB, Logistic, level
RandomTree