Institutional Repository of Vilnius University
Emocijų klasifikavimas lietuvių kalbos tekstuose taikant natūralios kalbos apdorojimo ir mašininio mokymosi metodus /
Abstract
dc:descriptionThis paper examines the task of emotion classification in Lithuanian text data by analyzing various text representation methods, including vectorization methods and linguistic features, and by applying classical machine learning algorithms, ensemble models, and a model based on the Trans former architecture. The study compared linguistic features, Bag-of-words, TFIDF, and Doc2Vec representations, as well as their combinations with linguistic features. For classification, we applied multinomial logistic regression, support vector machines, Naive Bayes, random forests, and extreme gradient boosting models. An ensemble model was also constructed, combining multiple logistic regression, support vector machines, and extreme gradient boosting algorithms, and the XLMRoBERTa transformer model was additionally evaluated. Experimental results showed that the best results were achieved using a combination of TF-IDF and linguistic features together with a support vector machine model, which demonstrated the highest accuracy among all the classical machine learning methods examined (accuracy = 0.8027; F1macro = 0.8028). Good results were also achieved by applying a combination of BoW and linguistic features with multinomial logistic regression: this method ensured high classification accuracy. The highest classification accuracy was achieved using the XLMRoBERTa model, which achieved an accuracy of 0.8934 and an F1macro score of 0.8949 on the test set. The ensemble model, combining multinomial logistic regression, support vector machines, and extreme gradient boosting models, achieved an accuracy of 0.8013, but no significant improvement was observed compared to the best individual model. The results show that classical methods, enriched with linguistic features, achieve a classification accuracy of 0.80, while transformer models demonstrate better results in the emotion classification task. Analysis of the support vector machine model showed that the classification results for neutral and joy emotions were the best, while lower classification metric values were observed for anger and surprise emotions.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Černiauskas, Ignas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD308334379&prefLang=en_US