Ex-Paper Analysis is a final year capstone project built at Thapar Institute of Engineering and Technology, Patiala, under the mentorship of Dr. Shivani Sharma. Built by a team of 4 students, the tool helps students structure exam preparation by automatically analyzing previous year question papers to identify the most frequently asked topics.
The Problem Students preparing for competitive and university exams spend significant time manually sifting through previous papers to find important topics — a tedious, error-prone process. Ex-Paper Analysis automates this entirely.
How it Works The system supports two modes:
Google Search Mode — User uploads a question paper PDF. The system segments the paper using image preprocessing and Masked R-CNN, extracts question text via OCR (EasyOCR), and searches Google via API to return the top 3 reference links per question in a structured table.
Book Search Mode — User uploads both a question paper and a textbook PDF. Both documents go through the full NLP pipeline: normalization → tokenization → stop word removal → stemming → TF-IDF matrix generation. The paper's TF-IDF matrix is cross-multiplied against the book's to identify which book topics most closely correspond to each exam question. Results are ranked by frequency and displayed with a pie chart breakdown. An Excel report is generated for download.
Technical Pipeline
- Masked R-CNN for question region detection and segmentation from scanned paper images
- OCR (EasyOCR/Textract) for text extraction from segmented question images
- TF-IDF for feature extraction and topic-question similarity scoring
- BERT model (explored for improved NLP accuracy over TF-IDF baseline)
- Google Search API for reference link retrieval
- React.js frontend, Django backend
Institution: Thapar Institute of Engineering and Technology, Patiala Team: 4 members | Mentor: Dr. Shivani Sharma