Auto-generation of Class Diagram from Free-text Functional Specifications and Domain Ontology

In this paper, we propose and implement a system that automates the building of class diagram from free-text requirement documents. Our approach first applies natural language processing (NLP) techniques to understanding of the written requirements, and then uses domain knowledge represented by domain ontology to improve the performance of class identification. The basic notion of our approach is that, a class especially core class of the domain is always semantically connected with other classes and its attributes. Based upon this, our system applies a spider model to search classes of interest. It first identifies core classes of the domain as a starting point, with which we are usually highly confident, and further finds classes that are semantically related with the identified ones. The strength of this method is that it identifies classes and relationships at one step. Specifically, our approach aims at addressing the following challenging research questions: (1) How to use domain ontology but not limited to domain ontology? (2) How to find candidate concepts? (3) How to find concept pairs with strong semantic connection in the context? (4) How to distinguish attribute name from class name for each concept? (5) How to name each inter-class relationship, aggregation, generalization, or association? (6) Are there any missing classes or attributes? This methodology finds its way through NLP, which is proved to be effective and sufficient to provide solutions to these problems in this paper.