Authors:
Gayane Sedrakyan
1
;
2
;
Asad Abdi
2
;
Stéphanie M. Van Den Berg
1
;
Bernard Veldkamp
1
and
Jos Van Hillegersberg
2
Affiliations:
1
Faculty of Behavioral, Management and Social Sciences, Section Cognition, Data & Education (CODE), University of Twente, Enschede, The Netherlands
;
2
Faculty of Behavioral, Management and Social Sciences, Section Industrial Engineering and Business Information Systems (IEBIS), University of Twente, Enschede, The Netherlands
Keyword(s):
Requirements Engineering, Requirements Analysis, Conceptual Modeling, Text Mining, Natural Language Processing, Requirement Analysis Automation, Model Generation.
Abstract:
Requirements analysis and modeling is a challenging task involving complex knowledge of the domain to be engineered, modeling notation, modelling knowledge, etc. When constructing architectural artefacts experts rely largely on the tacit knowledge that they have built based on previous experiences. Such implicit knowledge is difficult to teach to novices, and the cost of the gap between classroom knowledge and real business situations is thus reflected in further needs for post-graduate extensive trainings for novice and junior analysts. This research aims to explore the state-of-the art natural language processing techniques that can be adopted in the domain of requirements engineering to assist novices in their task of knowledge construction when learning requirements analysis and modeling. The outcome includes a method called Text-To-Model (TeToMo) that combines the state-of-the-art natural language processing approaches and techniques for identifying potential architecture elemen
t candidates out of textual descriptions (business requirements). A subsequent prototype is implemented that can assist a knowledge construction process through (semi-) automatic generation and validation of Unified Modeling Lnaguage (UML) models. In addition, to the best of our knowledge, a method that integrates machine learning based method has not been thoroughly studied for solving requirements analysis and modeling problem. The results of this study suggest that integrating machine learning methods, word embedding, heuristic rules, statistical and linguistic knowledge can result in increased number of automated detection of model constructs and thus also better semantic quality of outcome models.
(More)