Page 136 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 5 Depression discovery in cancer communities using deep learning 125

2.1Lexicon-based approaches
Lexicon-based methods for SA are very popular and widely
used [6e9]. In lexicon-based approaches, SA is carried out making
use of a preexisting lexicon, i.e., a dictionary of positive and nega-
tive words or by curating a sentiment lexicon for this task. Senti-
ment lexicons are curated using the manual approach,
dictionary-based approach, or corpus-based approach. The
manual approach is extremely monotonous and onerous. There-
fore, it is hardly used. Dictionary-based [10e13] and corpus-based
lexicon compilation approaches [14e17] are generally adopted.
Some of the works that are used to carry out SA using lexicons
are enlisted in Table 5.1. Authors in Ref. [18] use a hierarchical
lexicon for SA of news articles. In Ref. [19], the authors use the
popular sentiment lexicon SentiWordNet for carrying out SA on
movie reviews extracted from the movie review website Internet
Movie Database (IMDB). In this chapter, the authors use
discourse analysis to split the entire text into related and nonre-
lated and important and not important sections. They further
use weighing of sentiment scores in accordance with text impor-
tance to obtain the final sentiment scores of the movie reviews
under analysis. The experimental results obtained reinforce the
fact that splitting a text into important and nonimportant sections
and assigning weighted sentiment scores accordingly improves
the efficacy of SA [20]. However, this approach is computationally
expensive due to the large amount of time required in identifying
and segregating the important and nonimportant text segments
from one another. In Ref. [21], the authors use a combination of
sentiment lexicons for carrying out SA on the phrase level. They
integrate the polarity scores from all the lexicons using Markov
logic, neighborhood, and discourse relations. In Ref. [22], the au-
thors carry out lexicon-based SA on tweets. Authors in Ref. [23]
explore the prospect of using text streams as an alternative to
routine voting. They correlate public opinion identified from elec-
tions with the opinions expressed on Twitter. They use the Opin-
ionFinder Lexicon [24] and discover several correlations as well as
mine important trends. However, their investigations make them
realize the clear need for a more well-suited lexicon, incorporating
Twitter-specific features.
In Ref. [25], the authors analyze the sentiment in movie re-
views using SentiWordNet [26] and a domain-specific lexicon.
They compute the opinion polarities as well as the opinion
strength for each feature. However, their approach cannot handle
latent aspects, that is, aspects that may not be explicitly
mentioned in text.