Page 145 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

134 Chapter 5 Depression discovery in cancer communities using deep learning

They further show that incorporating the effect of adjacent words
increases the efficiency of SA. There are several limitations of this
work. Firstly, current window size for context is the whole docu-
ment, which needs refinement. Second, there needs to be a strat-
egy for smart filtering of useless context terms to reduce the size of
the contextualized lexicon and improve its quality. Thirdly, this
approach takes into account context only for ambiguous terms
and not for all the terms in the document.
In Ref. [52], the authors carry out SA on Twitter using POS-
specific prior polarity attributes. They investigate using a tree
kernel for an efficient tree representation of tweets to enable the
amalgamation of several types of attributes appropriately. The
major drawbacks of their approach are that parsing, SA, and topic
modeling are not considered.
Authors in Ref. [53] carry out SA on tweets using the Weka data
mining APIs. First, they use a sentiment lexicon and then research
in what manner to train the classifiers and define which is better
for tweet classification. The main limitations of their approach are
no enrichment of classification rules and ignoring of the SVM clas-
sifier, which is generally very effective in text processing and SA
applications.
In Ref. [74], the author has proposed an approach that solves
the problem of k-mean clustering and supervised ML algorithm
for the classification problem. The main problem of k-means is
to identify the value of “k,” which the author has sorted with lex-
ical chain approach that helped in finding the total number of
broad categories found in the given dataset. Based on the value
of “k,” the clustered data set is passed for the classification using
supervised algorithm that actually helps in improving the super-
vised algorithms in terms of time efficiency and accuracy both.
Such model falls under the category of semisupervised models.

2.4Other techniques
These constitute of approaches which are neither strictly ML
based nor lexicon based. Formal concept analysis (FCA) is one
such methodology. Fuzzy formal concept analysis (FFCA) is an
FCA variant that deals with ambiguity and vague facts and fig-
ures. FFCA has been effectively used in the information field in
the past [54].
Authors in Ref. [55] propose the use of FCA and FFCA for SA of
movies and e-book reviews corpora. They propose an approach
that utilizes FFCA to visualize the reviews as concepts. They use
training instances to decrease the random results triggered by
vague and confusing phrases. Using FFCA, they train a classifier