Increasing the semantic similarity of object-oriented domain models by performing behavioral analysis first

The main goal of any object-oriented analysis (OOA) method is to clarify a problem by modeling the problem and its domain. Therefore, the most important artifact that results from OOA is the domain model, which is usually realized as a class diagram that describes the core concepts in the domain and their relationships. Ideally, a mature engineering process is repeatable: analysts given the same problem and instructions to follow the same OOA process should produce semantically similar domain models. This work compares the observed semantic similarity among the different domain models pro-duced by one process for one system by different users of the process when the process is one of: 1. creation of use cases (UCs), then sequence diagrams, then a domain model, and 2. creation of UCs, then a unified UC statechart, then a domain model. One process was used to produce 31 specifications of a large VoIP system and its accompanying information management system. The other process was used to produce 34 specifications of the same system. The data show that domain models produced using the second process were 10% more semantically similar to each other than those produced using the first process, but at a cost, by one measure, of up to 25 % more time, spent in learning the process and in requirements elicitation. 1