Automated Handwritten Answer Script Assessment Using Deep Learning: A Synoptic Evaluation Framework

This research presented a comprehensive automated system for evaluating handwritten answer scripts through an innova-tive synoptic assessment methodology prioritizing conceptual understanding over surface-level textual similarity. The system integrates state-of-the-art optical character recognition (Mis-tral OCR), advanced vision-language modeling (Qwen2.5-VL-32B-Instruct), and modern natural language processing (Mistral 7B) to enable end-to-end automated evaluation of diverse response types including extended written answers and labeled diagrams. Extensive experimental validation on authentic examination scripts from Sardar Patel Institute of Technology demonstrated strong performance across multiple assessment dimensions. The system achieved 86.79% overall accuracy, 98% seman-tic accuracy, and 92.4% agreement with human evaluators, meeting or exceeding established performance targets and approaching inter-rater reliability typical among trained human evaluators. The multi-dimensional scoring framework provides compre-hensive assessment addressing semantic accuracy, key points coverage, factual correctness, and understanding depth. This granular evaluation offers more actionable feedback than sim-ple numeric scores and enables detailed performance analysis across different aspects of student understanding. Comparative benchmarking confirmed substantial advan-tages of the synoptic evaluation approach over traditional methods including keyword matching, edit distance similarity, and TF-IDF approaches. The semantic analysis capabilities of large language models enable recognition of paraphrased content and conceptually equivalent expressions that surface-level methods fail to identify.

[1]             S. Praveen, “An Approach to Evaluate Subjective Questions for Online Examination System,” International Journal of Innovative Research in Computer and Communication Engineering, vol. 2, pp. 6410–6413, 2014. [2]             A. Tulaskar, A. Thengal, and K. Koyande, “Subjective Answer Eval-uation System,” International Journal of Engineering Science and Computing, vol. 7, no. 4, pp. 10457–10459, 2017. [3]             H. Liu, C. Li, Y. Li, and Y. J. Lee, “Improved Baselines with Visual Instruction Tuning,” arXiv preprint arXiv:2310.03744, 2023.   A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. de las Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, L. R. Lavaud, M.-A. Lachaux, P. Stock, T. Le Scao, T. Lavril, T. Wang, T. Lacroix, and W. El Sayed, “Mistral 7B,” arXiv preprint arXiv:2310.06825, 2023. [4]             S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang, S. Wang, J. Tang, H. Zhong, Y. Zhu, M. Yang, Z. Li, J. Wan, P. Wang, W. Ding, Z. Fu, Y. Xu, J. Ye, X. Zhang, T. Xie, Z. Cheng, H. Zhang, Z. Yang, H. Xu, and J. Lin, “Qwen2.5-VL Technical Report,” arXiv preprint arXiv:2502.13923, 2025. [5]             I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning. MIT Press, 2016. [6]             A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is All You Need,” in Advances in Neural Information Processing Systems, 2017, pp. 5998–6008. [7]             J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in Proceedings of NAACL-HLT, 2019, pp. 4171–4186. [8]             T. B. Brown et al., “Language Models are Few-Shot Learners,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 1877–1901. [9]             A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning Transferable Visual Models From Natural Language Supervi-sion,” in Proceedings of the 38th International Conference on Machine Learning, 2021, pp. 8748–8763.