A lexicon based method to search for extreme opinions

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Sentiment analysis typically focuses on classifying opinions as positive, negative, or neutral.
• The identification of extreme opinions (most positive and most negative) has been largely overlooked in existing research.
• Extreme opinions hold significant value in various applications, yet lack dedicated analytical methods.

Purpose of the Study:

• To address the gap in sentiment analysis by developing a method for identifying extreme opinions.
• To propose an unsupervised approach for the automatic detection of highly polarized opinions.
• To construct a specialized lexicon for extreme sentiment classification.

Main Methods:

• An unsupervised learning approach was employed.
• A novel lexicon was automatically constructed.
• The lexicon specifically targets words expressing the most negative and most positive sentiments.

Main Results:

• Successfully developed an unsupervised method for extreme opinion identification.
• Created a lexicon containing words with extreme positive and negative polarity.
• Demonstrated the feasibility of identifying highly polarized opinions without labeled data.

Conclusions:

• The proposed unsupervised method effectively identifies extreme opinions.
• The automatically constructed lexicon is a valuable resource for specialized sentiment analysis.
• This research advances opinion mining by enabling the focus on the most impactful sentiments.