A Hybrid Lexicon-Machine Learning Approach to Macedonian Sentiment Analysis

Abstract

This study extends our previous work on a rule-based sentiment analysis system for Macedonian text [10], which relied on handcrafted lexicons and linguistic rules.We now investigate the integration of these rule-based features with supervised machine learning classifiers, specifically Logistic Regression (LR) and Support Vector Machines (SVM), to improve sentiment classification performance.Lexicon-derived features, including polarity, intensifiers, diminishers, and negation handling, are combined with statistical models to evaluate their impact.Experimental results show that the hybrid approach substantially outperforms the rule-based baseline, increasing the mean F1 score from 73.5% to 86.7% for SVM and 86.4% for LR.Paired t-tests confirm that these improvements are statistically significant (p < 0.001), while Wilcoxon tests indicate a strong trend (p = 0.0625).These findings demonstrate that integrating rule-based linguistic features with machine learning classifiers provides a robust framework for sentiment analysis in under-resourced languages such as Macedonian.

More by Jernej Vičič

• AI-generated text detection for Slavic languagesZenodo · 2026
• Dataset of Uzbek base words: extraction and data analysis based on the school corpusSocial Sciences & Humanities Open · 2026 · Citations: 2
• Dataset of Uzbek base wordsRepository of the University of Primorsk · 2026
• Uzbek 65 million web corpus lemmatized Word2Vec modelZenodo · 2026