Supervision : Marie-Pierre GERVAIS

Benefits of semi-supervised learning techniques in recovering traceability links between design artifacts

During the development of complex systems, several enterprises exchange a large number of heterogeneous models and requirements. During the phases of the system’s life cycle, these artifacts, linked to each other and derived from different modelling tools, are constantly evolving. In such a heterogeneous and volatile environment, it is necessary to manage the impact of the different changes occurring in the different design spaces. Traceability as defined by the International Council on Systems Engineering (INCOSE) meets this need.
However, establishing links between requirements and models in complex systems engineering requires dealing with a large volume of artifacts. For example, a specification of an autonomous vehicle with 3,000 requirements and 400 model elements, it would theoretically be necessary to check about one million of potential links. Although several approaches have been proposed for identifying traceability links, the validation process is always time-consuming and error-prone. This is mainly due to the predominance of manual operations during this process.
In this thesis, we propose a semi-supervised approach that learns through a probabilistic model to recognize links or no links from similarity measures and scores. This approach provides a quantitative confidence measure on each candidate link. This measure allows the expert in the validation phase to optimize his verification effort while reducing the risks of error. We evaluated our approach on benchmarks in the traceability and on industrial case studies. The results show that our approach have better results than state-of-the-art traceability methods. We obtain a reduction of no links (false positive) of about 80% compared to state-of-the-art methods in industrial cases, while, keeping a number of links (true positive), up to 75% , at the same time.
The prototype implemented, called Aggregation Trace Links Support (ATLaS), is being tested by the SystemX Research Institute’s partners as part of the Eco-mobility by Autonomous Vehicles (EVA) project.

2017-2019 Publications

• 2019
  • E. Effa Bella : “Apports des techniques d’apprentissage semi-supervisées dans l’établissement de liens entre artefacts de conception”, thesis, phd defence 10/28/2019, supervision Gervais, Marie-Pierre (2019)
  • E. EFFA BELLA, S. Creff, M.‑P. Gervais, R. Bendraou : “ATLaS: A Framework for Traceability Links Recovery Combining Information Retrieval and Semi-supervised Techniques”, 23RD IEEE INTERNATIONAL EDOC CONFERENCE - THE ENTERPRISE COMPUTING CONFERENCE, Paris, France (2019)
• 2018
  • E. EFFA BELLA, L. Wouters, M.‑P. Gervais, A. Koudri, R. Bendraou : “Semi-supervised Approach for Recovering Traceability Links in Complex Systems”, ICECCS 2018 - 23^rd International Conference on Engineering of Complex Computer Systems, Melbourne, Australia (2018)
• 2017
  • L. Wouters, S. Creff, E. EFFA BELLA, A. Koudri : “Towards Semantic-Aware Collaborations in Systems Engineering”, 24^th Asia-Pacific Software Engineering Conference (APSEC), Nanjing, China (2017)