Automatic Requirement Categorization of Large Natural Language Specifications at Mercedes-Benz for Review Improvements

摘要

Context and motivation: Today's industry specifications, in particular those of the automotive industry, are complex and voluminous. At Mercedes-Benz, a specification and its referenced documents often sums up to 3,000 pages. Question/problem: A common way to ensure the quality in such natural language specifications is technical review. Given such large specifications, reviewers have major problems in finding defects, especially consistency or completeness defects, between requirements with related information, spread over the various documents. Principal ideas/results: In this paper, we investigate two specifications from Mercedes-Benz, whether requirements with related information spread over many sections of many documents can be automatically classified and extracted using text classification algorithms to support reviewers with their work. We further research enhancements to improve these classifiers. The results of this work demonstrate that an automatic classification of requirements for multiple aspects is feasible with high accuracy. Contribution: In this paper, we show how an automatic classification of requirements can be used to improve the review process. We discuss the limitations and potentials of using this approach.