| In this thesis we focus on the issue of retrieving software from a collection of reusable software components. The proposed method is a variation of the relevance feedback process. The relevance feedback process is a popular retrieval mechanism typically used for test document environments. Here we describe how this method can be adapted for reusable software collections. Two methods of the relevance feedback process are presented. Also, experimental results of the methods' performance evaluation on two different software collections are presented and analyzed.; To the best of our knowledge, reusable software component retrieval is an area of study which has drawn little (or no) attention from the software engineering and information systems community. Most research reports on the technical aspects of software reusability deal with domain analysis, classification of software components, interoperability of software repositories, adaptation of software components for reuse, reuse of systems designs and architectures, and software metrics that quantify eligibility for reuse. All implicitly assume that the retrieval of software components is somehow taken care of automatically, or at best, it is done using some ad-hoc retrieval mechanism. The latter approach is typically quite ineffective since changes in the querying habits and/or modifications of the underlying software collection require major manual efforts for reorganization of the collection. Our proposal constitutes an attempt to put the software retrieval problem on the map of technical obstacles for software reuse. The proposed methods require no prior organization of the software collection. Instead, the collection is organized dynamically in response to the user's querying activity, until satisfactory values of recall and precision are achieved. With the exception of the phase in which the user has to relate her satisfaction of the retrieved results to the system, this organization is performed automatically and it is transparent to the user. |
