Kısa Yanıtlı Sınavlar İçin Yapay Zekâ Destekli Değerlendirme Sistemi

Öz

Kısa yanıtlı ve açık uçlu sınavların değerlendirilmesi, eğitim süreçlerinde yoğun zaman ve iş gücü gerektiren önemli problemlerden biridir. Geleneksel değerlendirme yöntemleri, özellikle kalabalık sınıflarda öğretim elemanları açısından ciddi iş yükü oluşturmakta ve öznel değerlendirme riskini artırmaktadır. Bu çalışmada, kısa yanıtlı sınavların otomatik olarak okunması, değerlendirilmesi ve sonuçlarının görselleştirilmesini sağlayan yapay zekâ destekli bütünleşik bir sistem geliştirilmiştir. Sistem; OCR tabanlı puan okuma, kesin cevaplı soruların otomatik değerlendirilmesi ve anlamsal benzerlik tabanlı açık uçlu cevap değerlendirmesi olmak üzere üç farklı senaryo üzerine kurulmuştur. Çalışmada istemci tarafında JavaFX tabanlı masaüstü uygulaması, sunucu tarafında ise Python ve FastAPI altyapısı kullanılmıştır. OCR performansını artırmak amacıyla görüntü ön işleme, karakter normalizasyonu ve Türkçe karakter düzeltme yöntemleri uygulanmıştır. Yapılan deneysel çalışmalar sonucunda OCR tabanlı puan okuma senaryosunda %82 başarı oranı elde edilmiş, kesin cevaplı değerlendirme senaryosunda karakter doğruluk oranı %99,50 seviyesine ulaşmaktadır. Açık uçlu değerlendirme senaryosunda ise sistem yaklaşık %78 doğruluk ve %82 F1-skoru elde etmiştir. Elde edilen sonuçlar, geliştirilen sistemin eğitim teknolojileri alanında uygulanabilir ve düşük maliyetli bir çözüm sunduğunu göstermektedir.

Anahtar Kelimeler

• Yapay Zekâ
• OCR
• Otomatik Sınav Değerlendirme
• Doğal Dil İşleme
• Eğitim Teknolojileri

Destekleyen Kurum

Bu çalışma TÜBİTAK 2209-A Üniversite Öğrencileri Araştırma Projeleri Destekleme Programı tarafından desteklenmektedir.

Etik Beyan

Bu çalışmanın tüm hazırlanma süreçlerinde etik kurallara uyulduğunu yazarlar beyan eder. Aksi bir durumun tespiti hâlinde Avrasya Üniversitesi Fen ve Mühendislik Bilimleri Dergisi’nin hiçbir sorumluluğu olmayıp, tüm sorumluluk çalışmanın yazarlarına aittir.

Teşekkür

Bu çalışma TÜBİTAK 2209-A Üniversite Öğrencileri Araştırma Projeleri Destekleme Programı tarafından desteklenmektedir.

Artificial Intelligence Supported Assessment System for Short Answer Exams

Öz

Assessment of short-answer and open-ended examinations is a time-consuming and labor-intensive process in educational environments. Traditional evaluation methods create significant workload for instructors, especially in crowded classrooms, and may lead to subjective grading inconsistencies. In this study, an integrated artificial intelligence-supported system was developed to automatically read, evaluate, and visualize short-answer examination results. The proposed system consists of three evaluation scenarios: OCR-based handwritten score extraction, automatic assessment of exact-answer questions, and semantic similarity-based evaluation of open-ended answers. A JavaFX-based desktop application was developed for the client side, while Python and FastAPI technologies were used on the server side. Image preprocessing, character normalization, and Turkish character correction techniques were implemented to improve OCR performance. Experimental results demonstrated an 82% success rate in OCR-based handwritten score extraction. As a result of the experimental studies conducted, an 82% success rate was achieved in the OCR-based score-reading scenario, and the character accuracy rate reached 99.50% in the binary classification scenario. In the semantic evaluation scenario, the system achieved approximately 78% accuracy and 82% F1-score. The findings indicate that the developed system provides a practical, low-cost, and applicable solution for educational technologies and automated assessment systems.

Anahtar Kelimeler

• Artificial Intelligence
• OCR
• Automatic Exam Assesment
• Natural Language Processing
• Education Technologies

Destekleyen Kurum

This study is supported by the TÜBİTAK 2209-A University Students Research Projects Support Program.

Etik Beyan

The authors declare that ethical rules are followed in all preparation processes of this study. In the case of a contrary situation, Avrasya University Science and Engineering Journal has no responsibility, and all responsibility belongs to the study's authors.

Teşekkür

This study is supported by the TÜBİTAK 2209-A University Students Research Projects Support Program.

Kaynakça

1. Anitha, R., Rajeev, R. R., Nazeem, M., & Navaneeth, S. (2024). Open-source OCR libraries: A comprehensive study for low resource language. International Centre for Free and Open Source Solutions (ICFOSS).
2. Bass, B. M. (1990). From transactional to transformational leadership: Learning to share the vision. Organizational Dynamics, 18(3), 19-31. https://doi.org/10.1016/0090-2616(90)90061-S
3. Badla, S. (2014). Improving the efficiency of Tesseract OCR engine [Master’s project, San Jose State University]. https://doi.org/10.31979/etd.5avd-kf2g
4. Bangare, S. L., Dubai, A. A., Bangare, S., & Patil, S. (2015). Reviewing Otsu’s method for image thresholding. International Journal of Applied Engineering Research, 10(9), 21777–21783. https://doi.org/10.37622/IJAER/10.9.2015.21777-21783
5. Bogachev, R. E., & Zarikovskaya, N. V. (2021). Overview of the capabilities of the FastAPI web framework for implementing the backend of web systems. In All-Russian Scientific Conference on Achievements of Science and Technology. https://doi.org/10.47813/dnit.2021.2.289-293
6. Bozkurt, F., Yağanoğlu, M., & Günay, F. B. (2015). Effective Gaussian blurring process on graphics processing unit with CUDA. International Journal of Machine Learning and Computing, 5(1), 57–61. https://doi.org/10.7763/IJMLC.2015.V5.483
7. Cui, C., Sun, T., Lin, M., Gao, T., Zhang, Y., Liu, J., Wang, X., Zhang, Z., Zhou, C., Liu, H., Zhang, Y., Lv, W., Huang, K., Zhang, Y., Zhang, J., Zhang, J., Liu, Y., Yu, D., & Ma, Y. (2025). PaddleOCR 3.0 technical report (Technical Report No. arXiv:2507.05595). arXiv. https://doi.org/10.48550/arXiv.2507.05595
8. Culjak, I., Abram, D., Pribanic, T., Dzapo, H., & Cifrek, M. (2012). A brief introduction to OpenCV. In Proceedings of the 35th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO) (pp. 1725–1730). IEEE. https://ieeexplore.ieee.org/document/6240859

9. Çelik, A. (2020). Optik karakter tanımada hata yayılım algoritmalarının performans kıyaslaması. Iğdır Üniversitesi Fen Bilimleri Enstitüsü Dergisi, 10(4), 2328–2340.
10. Duong, T. N. B., & Meng, C. Y. (2024). Automatic grading of short answers using large language models in software engineering courses. In 2024 IEEE Global Engineering Education Conference (EDUCON) (pp. 1-10). IEEE. https://doi.org/10.1109/EDUCON60312.2024.10578839
11. Er, Y. A., Kesen, İ., Şahin, G. G., & Erdem, A. (2025). Cetvel: A unified benchmark for evaluating language understanding, generation and cultural capacity of LLMs for Turkish. arXiv. https://arxiv.org/abs/2508.16431
12. Gemma Team. (2025). Gemma 3 technical report. ArXiv:2503.19786. https://doi.org/10.48550/arXiv.2503.19786
13. Jacob, M., Tshedza, M. P., & Hulisani, M. (2025). Tools, techniques, and applications of data visualization in education and machine learning. Research Square. https://doi.org/10.21203/rs.3.rs-7560670/v1
14. Li, M., Xu, Y., Cui, L., Huang, S., Wei, F., Li, Z., & Zhou, M. (2020). DocBank: A benchmark dataset for document layout analysis. arXiv. https://arxiv.org/abs/2006.01038
15. Li, Y. (2024). Synergizing optical character recognition: A comparative analysis and integration of Tesseract, Keras, Paddle, and Azure OCR [Master's thesis, University of Sydney]
16. Öztürk, M. G., Şahin, D. Ö., & Kılıç, E. (2025). Turkish optical character recognition under the lens: A systematic review of language-specific challenges, dataset scarcity, and open-source limitations. IEEE Access. Advance online publication. https://doi.org/10.1109/ACCESS.2025.3614147
17. Rahman, M. M., & Akter, F. (2019). An automated approach for answer script evaluation using natural language processing. International Journal of Computer Science, Engineering and Information Technology (IJCSET), 9, 39–47.
18. Rahman, M. M., & Siddiqui, F. H. (2018). NLP-based automatic answer script evaluation. DUET Journal, 4(1).
19. Rao, M. V., Sri Harshitha, I., Sukruthi, Y., & Sudharshan, T. (2020). Automatic answer script evaluator. International Journal of Scientific & Technology Research, 9(01), 4023–4028.
20. Reddy, P. P., Shruthi, P. S., Himanshu, P., & Singh, T. (2024). License plate detection using YOLO v8 and performance evaluation of EasyOCR, PaddleOCR and Tesseract. In 2024 15th International Conference on Computing Communication and Networking Technologies (ICCCNT) (pp. 1–6). IEEE. https://doi.org/10.1109/ICCCNT61001.2024.10725878
21. Saravanan, K., Chew, C. C., Tay, K. G., Kek, S. L., & Huong, A. (2022). Exam marks summation app using Tesseract OCR in Python. International Journal of Integrated Engineering, 14(3), 102–110. https://doi.org/10.30880/ijie.2022.14.03.011
22. Tulu, C. N., Ozkaya, O., & Orhan, U. (2021). Automatic short answer grading with SemSpace sense vectors and MaLSTM. IEEE Access, 9, 24620–24633. https://doi.org/10.1109/ACCESS.2021.3054346
23. Vieira, C., Parsons, P., & Byrd, V. (2018). Visual learning analytics of educational data: A systematic literature review and research agenda. Computers & Education, 122, 119–135. https://doi.org/10.1016/j.compedu.2018.03.018
24. Villar, S. A., Torcida, S., & Acosta, G. G. (2017). Median filtering: A new insight. Journal of Mathematical Imaging and Vision, 58(1), 130–146. https://doi.org/10.1007/s10851-016-0694-0
25. Vranesic, A. (2009). JavaFX [Diploma thesis, University of Ljubljana, Faculty of Computer and Information Science]. http://eprints.fri.uni-lj.si/1040/1/Vrane%C5%A1i%C4%8D_A.%2DVS.pdf
26. Xie, W., Niu, J., Xue, C. J., & Guan, N. (2025). Grade like a human: Rethinking automated assessment with large language models. In Proceedings of the International Conference on Research in Adaptive and Convergent Systems (pp. 1-8). https://doi.org/10.1145/3769002.3769962
27. Xu, Y., Li, M., Cui, L., Huang, S., Wei, F., & Zhou, M. (2020). Layoutlm: Pre-training of text and layout for document image understanding. In Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining (pp. 1192-1200). https://doi.org/10.1145/3394486.3403172
28. Yeşilyurt, Y. E. (2025). Evaluation of optical character recognition accuracy of Claude Opus 4 in handwritten student compositions. Nevşehir Hacı Bektaş Veli Üniversitesi SBE Dergisi, Sosyal Bilimlerde Yapay Zeka: Kuram, Uygulama ve Gelecek Perspektifleri, 142- 156. https://doi.org/10.30783/nevsosbilen.1770463
29. Zhao, C. (2025). Automatic short answer grading at scale using large language models [Master’s thesis, University of Illinois Urbana-Champaign].