Page 137 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

126 Chapter 5 Depression discovery in cancer communities using deep learning

Table 5.1 Literature Summary of Lexicon-based methods for sentiment analysis.

Level of
Authors Lexicon used Data set/Source Polarity Language granularity
Moreo Taxonomy lexicon designed for News articles Pos/neg English Document level
et al. news analysis
[18]
Heerschop SentiWordNet Internet Movie Database Pos/neg English Document level
et al.
[19]
Zirn et al. SentiWordNet The multidomain Pos/neg English Subsentence
[21] Unigram Lexicon, Taboda and sentiment data set level
Grieve's Turney Adjective (amazon.com)
List
Wu and d Twitter Pos/neg English Sentence level
Ren [22]
Connor OpinionFinder Twitter Pos/neg English Sentence level
et al.
[23]
Thet et al. SentiWordNet, domain-specific Movie reviews Pos/neg, English Aspect level
[25] lexicon sentiment
strength
VK Singh SentiWordNet Movie reviews Pos/neg English Document and
et al. aspect
[27] levels
Nasukawa Novel SA lexicon Web pages of various Pos/neg and English Document and
and Yi domains neutral sentence
[28] levels

In Ref. [27] also, the authors use the SentiWordNet lexicon for a
detailed feature and document-level SA of movie reviews. They
explore the use of “adverb þ verb” and “adverb þ adjective” com-
binations for document-level SA. Feature-level SA is domain spe-
cific. Feature-wise sentiment scores are assigned for all the
features in the text. The mean of the sentiment scores of all the
features represents the final movie review score. Their approach
has two major limitations. Firstly, since the aspect-level imple-
mentation is domain specific, there is a need to change the aspect
vector for different domains. Secondly, this approach cannot
handle the latent or hidden aspects.