Lexicon-based approach to the Sentiment and Emotion Analysis of a text in Azerbaijani language
Date Issued
2022-04
Author(s)
Mammadnaghiyev, Kheybar
Abstract
Sentiment Analysis of a text and Emotion Analysis of a text in the Azerbaijani language has
been recently at the center of attention as a research topic. Although different approaches have been
applied to this topic, there are still many approaches exist that can be applied to it. In this research,
the lexicon-based approach is applied to both Sentiment and Emotion Analysis. Briefly, sentiment
analysis deal with whether a text is positive, negative, or neutral, while emotion analysis detects one
of the basic emotions. Note that, in this research basic emotions of Plutchik's Wheel of Emotions are
used in order to categorize emotions which are anger, anticipation, disgust, fear, surprise, trust,
sadness, and joy. This research paper used 24,000 lines of news in Azerbaijani for training and testing
the models and used 66,000 lines of newly created Azerbaijani word dictionary dataset in order to
apply a lexicon-based approach. From the implementation point of view, Decision Trees, Naïve
Bayes, Support Vector Machine, Neural Network, and BERT machine learning algorithm are applied
and compared. Moreover, this research also deals with entity recognition. It detects the
corresponding sentiment and list of emotions of the main entity of the text. Note that, in this research,
entity recognition is accomplished using a rule-based approach.
been recently at the center of attention as a research topic. Although different approaches have been
applied to this topic, there are still many approaches exist that can be applied to it. In this research,
the lexicon-based approach is applied to both Sentiment and Emotion Analysis. Briefly, sentiment
analysis deal with whether a text is positive, negative, or neutral, while emotion analysis detects one
of the basic emotions. Note that, in this research basic emotions of Plutchik's Wheel of Emotions are
used in order to categorize emotions which are anger, anticipation, disgust, fear, surprise, trust,
sadness, and joy. This research paper used 24,000 lines of news in Azerbaijani for training and testing
the models and used 66,000 lines of newly created Azerbaijani word dictionary dataset in order to
apply a lexicon-based approach. From the implementation point of view, Decision Trees, Naïve
Bayes, Support Vector Machine, Neural Network, and BERT machine learning algorithm are applied
and compared. Moreover, this research also deals with entity recognition. It detects the
corresponding sentiment and list of emotions of the main entity of the text. Note that, in this research,
entity recognition is accomplished using a rule-based approach.