Feature Interaction Filtering with Use Case Maps at Requirements Stage

This paper proposes a new Feature Interaction (FI) ltering method at the requirements level. FI ltering is to screen out some irrelevant feature combinations before the FI detection process, by using simple indications that certain combinations are likely to cause FI. To achieve this, we extensively utilize a requirement notation method, called Use Case Maps (UCMs), which helps designers to visualize a global picture of call scenarios. Individual features are added to the basic call by using the stub plug-in concept of UCMs. That is, a set of UCMs describing the feature's functionality are plugged into certain parts (called stubs) of the basic call scenario in a \root" UCM. Thus, each feature is characterized by the stub con guration. Then, we propose a composition method of di erent stub con gurations in terms of a matrix, called a stub con guration matrix. Finally we present an FI ltering method for a given combination, which gives a verdict: (a) FI occurs, (b) FI never occurs or (c) FI-prone. Experimental evaluation using examples taken from the recent FI detection contest shows that almost half of the feature combinations can be ltered without costly feature analysis. This suggests that the number of feature combinations to be analyzed with more expensive FI detection methods can be reduced to half, by using the proposed method as a front end of the detection process.