REQ-I: Automated Requirements Identifier
- Project
- 20023 SmartDelta
- Type
- New product
- Description
Identifying requirements in large tender documents could aid the bidding process and help estimate the risk associated with the project. REQ-I formulates the requirement identification problem as a binary text classification problem. It uses various state-of-the-art classifiers based on traditional machine learning, deep learning, and few-shot learning for requirements identification in large tender documents. Results show that REQ-I could identify requirements in large documents with an average accuracy of 76%.
- Contact
- Mehrdad Saadatmand (RISE), Sarmad Bashir (RISE), Muhammad Abbas (RISE)
- {first.last}@ri.se
- Research area(s)
- NLP for requirements engineering
- Technical features
REQ-I identifies requirements in tender documents as follows:
- It extracts all the textual information from PDF tender documents using Optical Character Recognition (OCR)
- It queries a fine-tuned BERT large language model that classifies the text as either a requirement or not
- It then highlights the requirements in the PDF tender documents
- Integration constraints
Hugging Face Transformers, spaCy, NLTK, PyTorch, Numpy, Pandas, Tesseract
- Targeted customer(s)
Requirements Engineers, Bid Managers
- Conditions for reuse
Partly open-source: https://github.com/a66as/REFSQ2023-ReqORNot
- Confidentiality
- Public
- Publication date
- 15-11-2023
- Involved partners
- RISE - Research institutes of Sweden (SWE)
