| Buse Sarıçayır, Caner Özcan, Yasin Ortakçı A Comparative Study of Word Embedding Techniques for Turkish Sentiment Analysis |
|---|
| Abstract. This paper presents a comparative analysis of three different word embed-ding techniques—TF-IDF, Word2Vec, and FastText in the context of Turkish sentiment analysis. Each method offers a unique trade-off between computa-tional efficiency, representational richness, and classification accuracy of Turkish text. TF-IDF, a frequency-based approach, is computationally inex-pensive but neglects word order and contextual information. Word2Vec gen-erates context-aware embeddings capturing semantic relationships but re-mains context-independent for individual words. FastText further refines word representation by incorporating subword information, which is particu-larly advantageous for morphologically rich languages like Turkish. The per-formance of these embedding methods was evaluated using four different classification models (Logistic Regression, Decision Tree, Random Forest, and Support Vector Machine) on the Winvoker Turkish sentiment analysis dataset. Our experimental results show that Word2Vec achieves the highest accuracy with Logistic Regression. These results contribute to a better under-standing of the strengths and limitations of various word embedding tech-niques for sentiment analysis in morphologically rich languages. |
| Keywords: Sentiment Analysis, Turkish text, TF-IDF, Word2Vec, FastText, Embedding Models |
 Download PDF |
| DOI: https://doi.org/10.54381/itta2026.3.07 |