A classification methodology and retrieval model to support software reuse

Abstract

Studies have shown that reusing existing software can reduce development costs, speed up the development process, and provide a more reliable product. A software classification methodology and retrieval model have been developed to support the organization and location of reusable software components. This capability required the design and development of three cooperating processes: (1) an organization for the reusable software collection, (2) a method for describing software components, and (3) a mechanism to access (locate and retrieve) the desired software component. A faceted classification scheme and retrieval model were designed to overcome the deficiencies found in existing software reuse systems. These deficiencies include a lack of retrieval support for the developer, static classification schemes based on enumerative techniques which are difficult to expand or modify, and limited descriptive capabilities based on keyword retrieval technology. The retrieval model resulting from this research is one of the key components needed for large-scale software reuse. The faceted classification model from library science was used to design a software classification methodology based on an analysis and synthesis process. An analysis of the reusable software components is used to construct the classification. A synthesis process can then be used to describe items in the collection. The faceted methodology is adaptable to changes and growth in the target collection, concise in its descriptive format, facilitates automation, and supports citation order changes to adapt the organization to different users. Application of the faceted classification methodology to a test collection is presented. A formal retrieval model was designed using a combination of techniques from the boolean and vector space information retrieval models. The retrieval model uses attribute tuples to represent both the software components and user queries. The retrieval mechanism combines direct attribute matching from the boolean model with a similarity heuristic to provide relevance estimation. The general principles used to design the classification methodology and retrieval model were informally verified by experience with a rapid prototype system constructed as part of the research. Based on these initial experiences, the faceted classification methodology and hybrid retrieval mechanism appear to provide an effective retrieval system for reusable software components.