Türkçe Yazılmış Sahte Otel Yorumlarının Sınıflandırılması: Makine Öğrenmesi ve Derin Öğrenme Yöntemlerinin Karşılaştırmalı Analizi

Öz

Çevrimiçi otel değerlendirmeleri tüketicilerin karar verme süreçleri üzerinde önemli bir etkiye sahip olmakla birlikte, yanıltıcı ve sahte yorumların giderek yaygınlaşması bu değerlendirmelerin güvenilirliğini ciddi ölçüde tehdit etmektedir. Bu çalışma, etiketli veri setleri ve sınıflandırma çerçevelerinin sınırlı olduğu, biçimbilimsel açıdan zengin ve düşük kaynaklı bir dil olan Türkçede sahte yorumların sınıflandırılması problemini ele almaktadır. Araştırma kapsamında, TripAdvisor platformundan elde edilen 945 gerçek kullanıcı yorumu ile önceden eğitilmiş bir Türkçe GPT-2 dil modeli aracılığıyla sentetik olarak üretilen 944 yorum birleştirilerek toplam 1.889 otel yorumundan oluşan dengeli bir veri seti oluşturulmuştur. Çalışmada, Destek Vektör Makineleri (SVM), Rastgele Orman (Random Forest) ve CatBoost gibi geleneksel makine öğrenmesi algoritmalarının yanı sıra CNN, BiLSTM, DNN, CNN-BiLSTM ve CNN-BiLSTM-DNN mimarilerini içeren sekiz farklı sınıflandırma modeli sistematik olarak değerlendirilmiş ve karşılaştırılmıştır. Deneysel bulgular, hibrit derin öğrenme mimarilerinin hem tek başına kullanılan derin öğrenme modellerinden hem de geleneksel sınıflandırma algoritmalarından daha üstün performans sergilediğini ortaya koymuştur. Önerilen CNN-BiLSTM-DNN modeli, %93,39 doğruluk (accuracy), %93,85 kesinlik (precision), %93,61 F1 skoru ve %98,10 AUC değeri ile en yüksek genel performansa ulaşmıştır. Yazarların bilgisi dâhilinde, daha önce Türkçe GPT-2 modeli kullanılarak sentetik veri üretimi yoluyla sahte otel yorumlarının sınıflandırılmasına yönelik herhangi bir çalışma gerçekleştirilmemiştir. Bu yönüyle çalışma, biçimbilimsel açıdan zengin ve düşük kaynaklı diğer dillere de uygulanabilecek nitelikte özgün bir metodolojik katkı sunmaktadır.

Anahtar Kelimeler

• Türkçe doğal dil işleme
• sahte yorum sınıflandırması
• sentetik veri üretimi
• karar verme
• makine öğrenmesi.

Etik Beyan

Yazar tarafından herhangi bir çıkar çatışması beyan edilmemiştir. Bu çalışma etik kurul izni gerektirmemektedir.

Classification of Fake Hotel Reviews Written in Turkish: A Comparative Analysis of Machine Learning and Deep Learning Methods

Öz

Online hotel reviews significantly influence consumer decision-making, yet the growing prevalence of deceptive reviews poses a serious threat to their reliability. This study addresses the challenge of fake review classification in Turkish, a morphologically rich and low-resource language for which labeled datasets and classification frameworks remain scarce. A balanced dataset of 1,889 Turkish hotel reviews was constructed by combining 945 genuine TripAdvisor reviews with 944 synthetically generated reviews produced via a pre-trained Turkish GPT-2 language model. Eight classification models were systematically evaluated and compared, encompassing traditional machine learning algorithms, including SVM, Random Forest, and CatBoost, and deep learning architectures, including CNN, BiLSTM, DNN, CNN-BiLSTM, and CNN-BiLSTM-DNN. Experimental results demonstrated that hybrid deep learning architectures consistently outperformed standalone models and traditional classifiers. The proposed CNN-BiLSTM-DNN model achieved the highest overall performance, with an accuracy of 0.9339, precision of 0.9385, F1-score of 0.9361, and AUC of 0.9810. As far as the authors are aware, no prior work has attempted to apply synthetic data generation via a Turkish GPT-2 model for fake hotel review classification. This approach represents a methodological contribution with broad applicability to other morphologically rich, low-resource languages.

Anahtar Kelimeler

• Turkish natural language processing
• fake review classification
• synthetic data generation
• decision making
• machine learning

Etik Beyan

The authors have no conflict of interest to declare. This study does not require ethics board approval.

Kaynakça

1. Altork, Y. (2025). Comparative analysis of machine learning models for wind speed forecasting: Support vector machines, fine tree, and linear regression approaches. International Journal of Thermofluids, 27, 101217. https://doi.org/10.1016/j.ijft.2025.101217
2. Ansari, I., Hassani, K., Malakouti, S. M., & Suratgar, A. A. (2025). Predicting electrical energy consumption for finland using CNN-LSTM hybrid model. Next Research, 2(3), 100580. https://doi.org/10.1016/j.nexres.2025.100580
3. Ashraf, S., Rehman, F., Sharif, H., Kirn, H., Arshad, H., & Manzoor, H. (2023, 4–5 Jan. 2023). Fake Reviews Classification using Deep Learning 2023 International Multi-disciplinary Conference in Emerging Research Trends (IMCERT), https://doi.org/10.1109/IMCERT57083.2023.10075156
4. Aula, S. A., & Rashid, T. A. (2025). Foxtsage vs. Adam: Revolution or evolution in optimization? Cognitive Systems Research, 92, 101373. https://doi.org/10.1016/j.cogsys.2025.101373
5. Aytaç, Z. İ., İşeri, İ., & Dandil, B. (2025). A hybrid coot based CNN model for thyroid cancer detection. Computer Standards & Interfaces, 94, 104018. https://doi.org/10.1016/j.csi.2025.104018
6. Bergamin, L., & Aiolli, F. (2025). An investigation into creating counterfactual examples for non-linear Support Vector Machines. Neurocomputing, 651, 130809.
7. Bozuyla, M. (2022). Advanced Turkish fake news prediction with bidirectional encoder representations from transformers. Konya Journal of Engineering Sciences, 10(3), 750–761.
8. Budhi, G. S., Chiong, R., & Wang, Z. (2021). Resampling imbalanced data to detect fake reviews using machine learning classifiers and textual-based features. Multimedia Tools and Applications, 80(9), 13079–13097. https://doi.org/10.1007/s11042-020-10299-5

9. Cheng, M.-Y., Vu, Q.-T., & Mubarok, H. (2026). Satellite-inspired deep semantic mining for classifying contractual risk clauses. Expert Systems with Applications, 295, 128933.
10. Cui, J., Lv, C., Qu, X., Du, J., & Wang, H. (2025). Development of an intelligent CNN-LSTM-attention model for acoustic emission-based fracture detection and structural health monitoring in marine steel structures. Ocean Engineering, 339, 122002.
11. Çöltekin, Ç., Doğruöz, A. S., & Çetinoğlu, Ö. (2023). Resources for Turkish natural language processing: A critical survey. Language Resources and Evaluation, 57(1), 449–488.
12. Dong, S., Fang, Y., Shen, X., & Li, P. (2025). Design method of aeroengine turbine tenon/mortise joints’ teeth number based on active learning random forest and structural design optimization. Aerospace Science and Technology, 165, 110467. https://doi.org/10.1016/j.ast.2025.110467
13. Drir, N., Chekired, F., Mellit, A., & Blasuttigh, N. (2025). Hybrid CNN-EML model for fault diagnosis in electroluminescence images of photovoltaic cells. Renewable Energy, 250, 123343. https://doi.org/10.1016/j.renene.2025.123343
14. Fan, Z., Gou, J., & Weng, S. (2025). Complementary CatBoost based on residual error for student performance prediction. Pattern Recognition, 161, 111265.
15. Filieri, R., Alguezaui, S., & McLeay, F. (2015). Why do travelers trust TripAdvisor? Antecedents of trust towards consumer-generated media and its influence on recommendation adoption and word of mouth. Tourism Management, 51, 174–185. https://doi.org/10.1016/j.tourman.2015.05.007
16. Filieri, R., Lin, Z., Pino, G., Alguezaui, S., & Inversini, A. (2021). The role of visual cues in eWOM on consumers’ behavioral intention and decisions. Journal of Business Research, 135, 663–675. https://doi.org/10.1016/j.jbusres.2021.06.055
17. Gupta, D., Bhargava, A., Agarwal, D., Alsharif, M. H., Uthansakul, P., Uthansakul, M., & Aly, A. A. (2024). Deep Learning-Based Truthful and Deceptive Hotel Reviews. Sustainability, 16(11), 4514. https://doi.org/10.3390/su16114514
18. Hu, J., Huang, L., & Shao, Q. (2025). Combination models of random forest for predicting seismic liquefaction based on SPT, CPT, Vs databases considering sampling strategies. Soil Dynamics and Earthquake Engineering, 198, 109642. https://doi.org/10.1016/j.soildyn.2025.109642
19. Huang, W., Wang, Y., He, W., & Zhang, X. (2025). RTI-net: A decision support system for fish stress classification using multimodal learning network. Expert Systems with Applications, 276, 127174. https://doi.org/10.1016/j.eswa.2025.127174
20. Karadayı Ataş, P., Tak, N., Özöğür-Akyüz, S., & Eygi Erdogan, B. (2025). An innovative approach to ensemble learning in bankruptcy prediction using support vector machines and meta fuzzy functions. Information Sciences, 719, 122450. https://doi.org/10.1016/j.ins.2025.122450
21. Kim, M.-S., Park, S.-M., & Lee, J.-H. (2025). Development and validation of a DNN-based cooling water temperature optimization algorithm to achieve energy cost saving. Case Studies in Thermal Engineering, 73, 106697. https://doi.org/10.1016/j.csite.2025.106697
22. Li, K., Xie, X., Zhai, J., Zhou, B., Huang, C., & Wang, C. (2025). Tunnel lining defects identification using TPE-CatBoost algorithm with GPR data: A model test study. Tunnelling and Underground Space Technology, 157, 106275. https://doi.org/10.1016/j.tust.2024.106275
23. Li, R., Pan, C., Song, R., Liu, W., Xia, H., Teng, K., Zhou, C., Shang, C., & Hou, Y. (2025). Light-ShmooNet – A lightweight convolutional neural network for automated classification of shmoo plot with high accuracy. Expert Systems with Applications, 294, 128707. https://doi.org/10.1016/j.eswa.2025.128707
24. Lim, J.-S., Cho, H., Kwon, D.-H., & Lee, G.-S. (2025). Bi-LSTM based fault diagnosis scheme having high accuracy for Medium-Voltage Direct Current systems using pre- and post-processing. International Journal of Electrical Power & Energy Systems, 169, 110793. https://doi.org/10.1016/j.ijepes.2025.110793
25. Lima, S. F., Portmann, E., & Terán, L. (2025). FuzzyArcLoss: Dynamic margin adjustment for robust recognition across domains. Expert Systems with Applications, 281, 127477. https://doi.org/10.1016/j.eswa.2025.127477
26. Lu, J., Zhan, X., Liu, G., Zhan, X., & Deng, X. (2023). BSTC: A Fake Review Detection Model Based on a Pre-Trained Language Model and Convolutional Neural Network. Electronics, 12(10), 2165. https://doi.org/10.3390/electronics12102165
27. Luo, T., Qu, J., & Cheng, S. (2025). Technological opportunity discovery based on VERGM and random forest model. Expert Systems with Applications, 293, 128712.
28. Madhuri, J., Ghana Shyam, S., Kumar, A. N., Gagan, D., & Rohit, A. (2024, 17–19 Oct. 2024). Enhanced Fake Review Detection and Sentiment Analysis with Hybrid Deep Learning Models 2024 IEEE International Conference on Blockchain and Distributed Systems Security (ICBDS), https://doi.org/10.1109/ICBDS61829.2024.10837222
29. Minh, T. Q., Ngo Van, T., Nguyen, H. X., & Nguyễn, Q. (2025). Enhancing the Structural Health Monitoring (SHM) through data reconstruction: Integrating 1D convolutional neural networks (1DCNN) with bidirectional long short-term memory networks (Bi-LSTM). Engineering Structures, 340, 120767. https://doi.org/10.1016/j.engstruct.2025.120767
30. Mohawesh, R., Salameh, H. B., Jararweh, Y., Alkhalaileh, M., & Maqsood, S. (2024). Fake review detection using transformer-based enhanced LSTM and RoBERTa. International Journal of Cognitive Computing in Engineering, 5, 250–258. https://doi.org/10.1016/j.ijcce.2024.06.001
31. Najafabadi, M. K. (2024). Sentiment analysis incorporating convolutional neural network into hidden Markov model. Computational Intelligence, 40(2), e12633. https://doi.org/10.1111/coin.12633
32. Oflazer, K. (2014). Turkish and its challenges for language processing. Language Resources and Evaluation, 48(4), 639–653. https://doi.org/10.1007/s10579-014-9267-2
33. Ott, M., Choi, Y., Cardie, C., & Hancock, J. T. (2011, June). Finding Deceptive Opinion Spam by Any Stretch of the Imagination Portland, Oregon, USA. https://aclanthology.org/P11-1032/
34. Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language models are unsupervised multitask learners. OpenAI blog, 1(8), 9.
35. Reddy, R., & Kumar, U. M. A. (2022, 25–27 May 2022). Convolutional Neural Networks- Bidirectional Long Short Term Memory Based Fake Review Classification 2022 6th International Conference on Intelligent Computing and Control Systems (ICICCS), Shaikh, S., Daudpota, S. M., Imran, A. S., & Kastrati, Z. (2021). Towards Improved Classification Accuracy on Highly Imbalanced Text Dataset Using Deep Neural Language Models. Applied Sciences, 11(2), 869. https://doi.org/10.3390/app11020869
36. Sikchi, P., Jagtap, S., & Kundale, J. (2024, 5–7 March 2024). Fake Review Detection using Neural Network 2024 International Conference on Emerging Smart Computing and Informatics (ESCI), https://doi.org/10.1109/esci59607.2024.10497283
37. Singh, R., & Sharma, A. (2025). STAD-ConvBi-LSTM: Spatio-temporal attention-based deep convolutional Bi-LSTM framework for abnormal activity recognition. Journal of Visual Communication and Image Representation, 110, 104465.
38. So, B., & Valdez, E. A. (2025). Zero-inflated Tweedie boosted trees with CatBoost for Insurance loss analytics. Applied Soft Computing, 177, 113226. https://doi.org/10.1016/j.asoc.2025.113226
39. Sun, P., Bi, W., Zhang, Y., Wang, Q., Kou, F., Lu, T., & Chen, J. (2024). Fake Review Detection Model Based on Comment Content and Review Behavior. Electronics, 13(21), 4322.
40. Tang, W., & Yang, Z. (2025). DNN-based Parameterization for B-Spline Curve Approximation. Computer-Aided Design, 186-187, 103897. https://doi.org/10.1016/j.cad.2025.103897
41. Thao, L. Q., Kien, D. T., Thuy, D. T. T., Thuy, L. T. M., Bach, N. C., Duc, T. T., Bach, H. G., & Cuong, D. D. (2024). Designing a deep learning-based application for detecting fake online reviews. Engineering Applications of Artificial Intelligence, 134, 108708.
42. Toprak Kesgin, H., Yuce, M. K., Dogan, E., Egemen Uzun, M., Uz, A., Seyrek, H. E., Zeer, A., & Amasyali, M. F. (2024). Introducing cosmosGPT: Monolingual Training for Turkish Language Models. arXiv e-prints, arXiv: 2404.17336. https://doi.org/10.48550/arXiv.2404.17336
43. Veda, C. R., Apoorva, N., Vishnu, N., Velpuru, M. S., Akshaya, H., & Siripuram, S. P. (2024, 8–9 Nov. 2024). Fake Review Identification using Hybrid Fusion of Machine Learning and Natural Language Processing Techniques 2024 3rd Odisha International Conference on Electrical Power Engineering, Communication and Computing Technology (ODICON),