EXISTING METHODOLOGY
The recognized text is cleaned, removing noise and handling text errors. Simple rule-based
systems use predefined keywords and phrases for evaluation. The answer is compared with the
ideal solution by looking for the presence of key terms. Points are awarded based on the presence
of keywords or the proximity of words in sentences. This is often rigid and lacks deep
comprehension.
Struggles with handwritten text.
Inability to understand meaning and context.
Difficulty in recognizing diverse handwriting styles and answer structures.
Lack of adaptive learning for continuous improvement.
PROPOSED METHODOLOGY
Use deep learning models like Convolutional Neural Networks (CNNs) for better recognition
of complex and diverse handwriting. Models like CRNN (Convolutional Recurrent Neural
Network) can be used to handle sequences of characters for handwritten text. Pre-trained
models such as Google Vision or Microsoft Azure’s OCR can be integrated for higher
accuracy. Using NLP libraries like spaCy or NLTK for automatic correction of OCR errors.
Identify key terms, names, or specific entities within the answer to understand structure and
focus. Utilize models like BERT (Bidirectional Encoder Representations from Transformers)
or GPT (Generative Pre-trained Transformer) for semantic understanding of the answers.
These models can analyze the context and meaning of the text rather than just keyword
matching. Employ sentence similarity algorithms to compare student answers with model
answers. Models like Universal Sentence Encoder or BERT embeddings can quantify how
closely the student's answer aligns with the ideal answer. Rather than keyword-based, the
proposed system can use context-based evaluation, where the AI understands the meaning of
an answer and compares it to the model answers. Use reinforcement learning or attention-
based models that can identify relevant portions of an answer and provide partial credit even
if the whole answer is not correct. The AI learns from human graders' feedback using deep
learning techniques, ensuring consistent and fair grading over time. NLP algorithms can
assess whether the tone of the answer matches what is expected in specific question types
(e.g., descriptive vs analytical). Integrate tools like Grammarly API or customized syntax
checking systems to grade based on language structure, grammar, and overall writing quality.
For generating model answers to compare with student responses. Improved Deep learning-
based OCR enhances text recognition accuracy, especially for handwritten scripts. NLP
models like BERT help in understanding the meaning behind answers, providing a more
accurate assessment. Reduces human intervention in grading, saving time and reducing
errors. The system becomes more reliable and unbiased over time through continuous
learning.
