Automatic classification of software requirements, a major component necessary in the analysis of requirements, has been the topic of extensive research efforts in recent years. An approach to automated classification that involves the application of indexing and clustering techniques is described here. A major objective of this process is to aggregate a set of N requirements into a set of M requirements clusters where M•N. A two-tiered clustering (TTC) algorithm is presented that enables the authors to discriminate between requirements statements and group them in accordance with similarity characteristics that exist across requirements statements. Once this aggregation is accomplished and requirements are classified through this indexing and clustering technique, it is possible to analyze them to determine the existence or absence of problems such as conflict, incompleteness, inconsistency, and imprecision within and across requirements statements clusters. A test case involving a very large set of requirements has been investigated and the results show that the TTC algorithm provides the information necessary to successfully differentiate between similar and dissimilar requirements and cluster similar requirements. This approach appears to be much more flexible than other indexing schemes such as the faceted approach and the predefined taxonometric approach.
