Abstract

The effectiveness of oxymorons in social media hinges on their context, target audience, and usage frequency. Using oxymorons can help clarify nuanced semantic variations or highlight inherent conflicts. The primary objective of this research endeavor is to develop a state-of-the-art oxymoron classifier. To achieve this, a comprehensive feature extraction process was undertaken, encompassing N-grams, part-of-speech tags, and structural features from a meticulously balanced dataset. These extracted features were then integrated with various feature weighting schemes and evaluated using a suite of machine learning algorithms, including Random Forest (RF), Naïve Bayes (NB), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN). The proposed KNN algorithm, when used in conjunction with all features and the TF-IDF weighting scheme, demonstrated superior classification performance, achieving precision, recall, F-measure, accuracy, and kappa scores of 99.38%, 99.64%, 99.51%, 99.50%, and 0.99, respectively. These results demonstrate the superior performance of the KNN classifier in the context of oxymoron classification. Future advancements in this research will focus on predicting oxymoronic phrases within mixed-language environments.

Authors

K. Seethappan¹, K. Premalatha²
Anna University Regional Campus, Coimbatore, India¹, Bannari Amman Institute of Technology, India²

Keywords

Machine learning, Sentiment Analysis, Natural Language Processing, Figurative Language, Oxymoron