Çok Etiketli Duygu Sınıflandırmasında Kelime Gömme Yöntemlerinin Derin Öğrenme Modelleri ile Kapsamlı Bir Değerlendirmesi

Öz

Bu çalışmada, duygu analizi alanında kullanılan farklı kelime gömme yöntemlerinin derin öğrenme modellerinin performansı üzerindeki etkileri incelenmiştir. Google tarafından geliştirilen GoEmotions veri kümesi kullanılarak çok etiketli metinler üzerinde duygu sınıflandırması gerçekleştirilmiştir. Deneysel süreçte GloVe, FastText ve Word2Vec olmak üzere üç farklı kelime gömme yöntemi uygulanmış ve elde edilen vektör temsilleri çeşitli derin öğrenme mimarilerine (CNN, RNN, LSTM, GRU) girdi olarak verilmiştir. Ayrıca, bu mimarilere dikkat (attention) mekanizması entegre edilerek, dikkat katmanının model performansı üzerindeki etkisi analiz edilmiştir. Elde edilen bulgular, kelime gömme yöntemlerinin duygu sınıflandırma modellerinin başarımı üzerinde anlamlı farklılıklar yarattığını göstermektedir. Özellikle, CNN+Attention mimarisi ile GloVe gömme yönteminin kullanıldığı senaryoda en yüksek kesinlik değeri (%71.60) elde edilmiştir. Öte yandan, en dengeli ve etkili performans, BiGRU+Attention mimarisi ile GloVe yönteminin birlikte kullanıldığı durumda %56.11’lik F1 skoru ile gözlemlenmiştir. Literatür incelemesi sonucunda, GoEmotions veri kümesi gibi çok sınıflı bir yapıya sahip veri seti üzerinde üç farklı gömme yönteminin farklı derin öğrenme mimarileri ile karşılaştırmalı olarak değerlendirildiği benzer bir çalışmaya rastlanmamıştır. Bu sonuçlar, duygu analizi uygulamalarında uygun kelime gömme yönteminin seçiminde dikkatli olunması gerektiğini ortaya koymakta ve kelime temsil teknikleri ile derin öğrenme mimarilerinin entegrasyonuna yönelik özgün katkılar sunmaktadır.

Anahtar Kelimeler

• Doğal Dil İşleme
• Metin Sınıflandırma
• Çok Etiketli Duygu Analizi
• Kelime Gömme
• Derin Öğrenme

A Comprehensive Evaluation of Word Embedding Methods with Deep Learning Models for Multi-Label Emotion Classification

Öz

This study investigates the effects of different word embedding methods on the performance of deep learning models in the field of emotion analysis. Using the GoEmotions dataset developed by Google, emotion classification was performed on multi-label texts. In the experimental procedure, three distinct word embedding methods — GloVe, FastText, and Word2Vec — were applied, and the resulting vector representations were fed into various deep learning architectures (CNN, RNN, LSTM, GRU). In addition, an attention mechanism was integrated into these architectures to analyze the impact of attention layers on model performance. The findings indicate that word embedding methods produce statistically meaningful differences in the success of emotion classification models. In particular, the highest precision was achieved (71.60%) in the scenario combining the CNN+Attention architecture with GloVe embeddings. On the other hand, the most balanced and effective overall performance was observed for the BiGRU+Attention architecture paired with GloVe, yielding an F1 score of 56.11%. A literature review revealed no prior study that comparatively evaluates three different embedding methods across multiple deep learning architectures on a multi-label dataset such as GoEmotions. These results underscore the need for careful selection of appropriate word embedding techniques in emotion analysis applications and provide original contributions to the integration of word representation techniques with deep learning architectures.

Anahtar Kelimeler

• Natural Language Processing
• Text Classification
• Multi-Label Emotion Analysis
• Word Embedding
• Deep Learning

Kaynakça

1. D. Demszky, D. Movshovitz-Attias, J. Ko, A. Cowen, G. Nemade, ve S. Ravi, “GoEmotions: A Dataset of Fine Grained Emotions”, 03 Haziran 2020, arXiv: arXiv:2005.00547. doi: 10.48550/arXiv.2005.00547.
2. V. Kumar, A. Bansal, ve U. Gupta, “Comparative Analysis of Text based Emotion Detection on GoEmotions Dataset”, içinde 2023 5th International Conference on Advances in Computing, Communication Control and Networking (ICAC3N), Ara. 2023, ss. 364-369. doi: 10.1109/ICAC3N60023.2023.10541692.
3. N. Birannavar, P. Prasad, G. Dandgal, M. Ganiger, P. Redekar, ve V. Badiger, “Performance Evaluation of Sentiment Analysis on Reddit Comments: Insights and Improvement Opportunities for Naive Bayes, SVM, and BERT Models”, içinde 2025 International Conference on Computer, Electrical & Communication Engineering (ICCECE), Şub. 2025, ss. 1-5. doi: 10.1109/ICCECE61355.2025.10940395.
4. A. Younas, S. Riaz, S. Ali, R. Khan, M. Ullah, ve D. Kwak, “Stacked LSTM Model for Contextual Correlation Detection Among Multiple Emotions”, IEEE Access, ss. 1-1, 2025, doi: 10.1109/ACCESS.2025.3582764.
5. N. Alvarez-Gonzalez, A. Kaltenbrunner, ve V. Gómez, “Uncovering the Limits of Text-based Emotion Detection”, 2021. doi: 10.18653/v1/2021.findings-emnlp.219.
6. J. Zhou, S. Luo, ve H. Chen, “Expansion Quantization Network: An Efficient Micro-emotion Annotation and Detection Framework”, 30 Mayıs 2025, arXiv: arXiv:2411.06160. doi: 10.48550/arXiv.2411.06160.
7. E. S. Yusifov ve I. S. Sineva, “An Intelligent System for Assessing the Emotional Connotation of Textual Statements”, içinde 2022 Wave Electronics and its Application in Information and Telecommunication Systems (WECONF), May. 2022, ss. 1-8. doi: 10.1109/WECONF55058.2022.9803516.
8. Y. Ni ve W. Ni, “A multi-label text sentiment analysis model based on sentiment correlation modeling”, Front. Psychol., c. 15, Ara. 2024, doi: 10.3389/fpsyg.2024.1490796.

9. J. Olah, S. Baruah, D. Bose, and S. Narayanan, “Cross Domain Emotion Recognition using Few Shot Knowledge Transfer,” Oct. 11, 2021, arXiv: arXiv:2110.05021. doi: 10.48550/arXiv.2110.05021.
10. A. Bhardwaj, N. A. Wasi, and M. Abulaish, “Unlocking Emotions in Text: A Fusion of Word Embedding and Lexical Knowledge for Emotion Classification”.
11. Y. Sazaki, A. F. Yusof, A. R. C. Hussin, and C. W. Howe, “Effect of Training Epochs on Multi-Label Emotion Classification Using RoBERTa with Focal Loss,” in 2025 19th International Conference on Telecommunication Systems, Services, and Applications (TSSA), Oct. 2025, pp. 1–4. doi: 10.1109/TSSA68467.2025.11303830.
12. M. M and S. Gokila, “Beyond Single Emotions: A Dual-Label Classification System for Enhanced Emotion Understanding in NLP Using Transformer-Based Models,” in 2025 8th International Conference on Trends in Electronics and Informatics (ICOEI), Apr. 2025, pp. 1491–1497. doi: 10.1109/ICOEI65986.2025.11013237.
13. Q. Peng, Y. Wan, and X. Zeng, “Multi-Label Emotion Recognition Model Integrating Multi-Perspective Common Sense Knowledge,” in 2025 7th International Conference on Natural Language Processing (ICNLP), Mar. 2025, pp. 421–425. doi: 10.1109/ICNLP65360.2025.11108467.
14. J. Pennington, R. Socher, and C. Manning, “GloVe: Global Vectors for Word Representation,” in Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), A. Moschitti, B. Pang, and W. Daelemans, Eds., Doha, Qatar: Association for Computational Linguistics, Oct. 2014, pp. 1532–1543. doi: 10.3115/v1/D14-1162.
15. P. Bojanowski, E. Grave, A. Joulin, and T. Mikolov, “Enriching Word Vectors with Subword Information,” Transactions of the Association for Computational Linguistics, vol. 5, pp. 135–146, 2017, doi: 10.1162/tacl_a_00051.
16. T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient Estimation of Word Representations in Vector Space,” Sept. 07, 2013, arXiv: arXiv:1301.3781. doi: 10.48550/arXiv.1301.3781.
17. S. K. Putri, A. Amalia, and T. F. Abidin, “Sentiment Analysis Multi-Label of Toxic Comments using BERT-BiLSTM Methods,” in 2024 International Conference on Electrical Engineering and Informatics (ICELTICs), Sept. 2024, pp. 120–124. doi: 10.1109/ICELTICs62730.2024.10776338.
18. D. Alnahas, F. Aşık, A. Kanturvardar, and A. M. Ülkgün, “Opinion Mining Using LSTM Networks Ensemble for Multi-class Sentiment Analysis in E-commerce,” in 2022 3rd International Informatics and Software Engineering Conference (IISEC), Dec. 2022, pp. 1–6. doi: 10.1109/IISEC56263.2022.9998264.
19. J. Chung, C. Gulcehre, K. Cho, and Y. Bengio, “Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling,” Dec. 11, 2014, arXiv: arXiv:1412.3555. doi: 10.48550/arXiv.1412.3555.
20. A. N. Ma’Aly, D. Pramesti, and H. Fakhrurroja, “Comparative Analysis of Deep Learning Models for Multi-label Sentiment Classification of 2024 Presidential Election Comments,” in 2024 7th International Conference on Informatics and Computational Sciences (ICICoS), July 2024, pp. 502–507. doi: 10.1109/ICICoS62600.2024.10636889.
21. A. Graves and J. Schmidhuber, “Framewise phoneme classification with bidirectional LSTM and other neural network architectures,” Neural Networks, vol. 18, no. 5, pp. 602–610, July 2005, doi: 10.1016/j.neunet.2005.06.042.
22. A. Vaswani et al., “Attention Is All You Need,” Aug. 02, 2023, arXiv: arXiv:1706.03762. doi: 10.48550/arXiv.1706.03762.
23. N. T. M. Trang and N. N. Hung, “BiGRU-Attention Sentiment Analysis for Enhancing Task-Oriented Chatbots,” in 2024 5th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI), Jan. 2024, pp. 476–483. doi: 10.1109/ICMCSI61536.2024.00075.
24. M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,” Information Processing & Management, vol. 45, no. 4, pp. 427–437, July 2009, doi: 10.1016/j.ipm.2009.03.002.
25. M. Coskun, “Enhancing Reliability in Deep Learning Diagnosis of Brain Tumors Using Grad-CAM,” TDFD, vol. 15, no. 2, pp. 39–48, Jul. 2026, doi: 10.46810/tdfd.1749282.
26. M. Grandini, E. Bagli, and G. Visani, “Metrics for Multi-Class Classification: an Overview,” Aug. 13, 2020, arXiv: arXiv:2008.05756. doi: 10.48550/arXiv.2008.05756.