SENTIMENT ANALYSIS OF TEXT REVIEWS USING TONE DICTIONARIES AND FUZZY SET CARDINALITY

Abstract

Sentiment or opinion analysis aims to determine the polarity of people's opinions in relation to any product, service, event or any person. One of the most common methods used in sentiment analysis of text content is natural language processing. Sentiment analysis of natural language text can be assessed using numerous methodologies such as machine learning algorithms and statistical tools, while the application of fuzzy logic is not common. The use of fuzzy logic was chosen for the following reasons. First, fuzzy logic handles linguistic uncertainty well. This way of defining the problem leads to a reduction in bias, both positively and negatively. Secondly, learn ing approaches based on fuzzy rules are fundamentally different from those learning approaches that are widely used in sentiment classification, such as support vector machines, naive Bayes, etc., as they relate to generative learning, i.e. i.e. the goal of learning is to assess the degree to which an instance belongs to each individual class. The proposed model for sentiment analysis of text reviews is based on the use of tone lexicons using fuzzy logic and consists of four main stages. The steps include tokenization, word bag model formulation, sentiment fuzzy score formulation, and polarity assignment. In the proposed model, the power of the fuzzy set is used as a measure of the evaluation of the indicators of the polarity of words. Word polarity values are obtained by applying two sentiment lexicons: SentiWordNet and AFINN. Two versions of the model were created depending on the type of vocabulary used: based on SentiWordNet and AFINN. Comparison of the presented approach based on fuzzy logic with other dictionary-based methods demonstrates the superiority of the developed models based on the application of fuzzy logic.