Context and challenges
This client is a major player in the nuclear industry.
The project launched by this client aimed to better manage the New Nuclear and Nuclear Power Plant projects through an in-depth update of the technical documentary repositories.
Today, the transformation of document repositories into data-centric repositories required a big human investment and needs the help of machine and deep learning technologies.
In this context, this client required Assystem to intervene in this project in order to help extracting, classifying and rationalizing requirements for more than 100 documents to build a strong and dynamic requirements database.
Deployment of DeepREXT solution (Advanced Natural Language Processing methods able to capture custom semantics):
- Phase 1: Requirements extraction
- Detect and extract requirements from a raw technical documentary
- Phase 2: Requirements classification
- Train the model to classify the extracted requirements according to a precise classification map given by the customer
- Automate the classification of requirements
- Phase 3: Requirements rationalization
- Automate the comparison of sets of requirements that have been produced by different entities to identify duplicated and inconsistent requirements
- Time savings through accelerating by at least 50% the transformation of requirements gathered from raw documents into a unified format thanks to automation of requirements extraction
- Time savings (no need to write custom rules) thanks to the training on thousands of labelled examples, enabling the models to be capable of capturing all necessary dependencies
- Smarter algorithms thanks to collection of feedback from engineers to adjust models (in particular feedback of requirements classification and extraction)
- Good understanding of various styles of writing of the different authors thanks to the flexibility of DeepREXT enabling to understand the underlying meaning of words of all requirements