SentiHealth: creating health-related sentiment lexicon using hybrid approach

Abstract

The exponential increase in the health-related online reviews has played a pivotal role in the development of sentiment analysis systems for extracting and analyzing user-generated health reviews about a drug or medication. The existing general purpose opinion lexicons, such as SentiWordNet has a limited coverage of health-related terms, creating problems for the development of health-based sentiment analysis applications. In this work, we present a hybrid approach to create health-related domain specific lexicon for the efficient classification and scoring of health-related users’ sentiments. The proposed approach is based on the bootstrapping modal, a dataset of health reviews, and corpus-based sentiment detection and scoring. In each of the iteration, vocabulary of the lexicon is updated automatically from an initial seed cache, irrelevant words are filtered, words are declared as medical or non-medical entries, and finally sentiment class and score is assigned to each of the word. The results obtained demonstrate the efficacy of the proposed technique.

Perspectives of the performance metrics in lexicon and hybrid based approaches: a review

“…The cosine similarity score W2Vr is assigned to r in the Word2Vec. The creation of health associated sentiment lexicon with hybrid approach are explained [21]. This methodology uses the bootstrapping, health opinions dataset and corpus based sentiment lexicon for preprocessing, creation and lexicon expansion, removing irrelevant words, polarity determination and polarity score modification.…”

“…The methodology proposed by [20] polaritysim, SVM, MNB is performed on restaurant data set which achieves highest accuracy of 91%. A methodology such as bootstrapping, corpus based polarity detection and scoring, which are performed on health care review data set achieve precision of 89%, recall of 79% and F-Measure of 83% [21]. A methodology termed TME, which achieves highest classification accuracy as 86.06%, Mean of average precision as 87% and Average Pearson's correlation as 48% are discussed in [19].…”

“…It requires to increase the classification of precision with ensemble methods and different dictionaries [13]. An approach based on bootstrapping, corpus based polarity detection and scoring for creating a health-related sentiment lexicon are presented [21]. However, it needs specific health-related domains to reduce the noise in the expanded lexicon.…”

Sentiment Analysis Tool for Pharmaceutical Industry & Healthcare

“…As defined in [19], sentiment analysis is a computational study of opinions, sentiments, Subjectivity, emotions, and attitude expressed in texts towards an entity. Thus, SA is a multitask of detecting, extracting and classifying opinions, sentiments and attitudes concerning different topics, as expressed in textual input.In Fig 1.presents seven broad dimensions of SA.…”

“…The SHCpt tool is based on the method called SentiHealth proposed in ther paper [19]. Moreover, as a second alternative is to analyze the texts in Portugues.…”