Intelligent classification and retrieval of software components

Abstract

This work proposes a new methodology for intelligent classification and retrieval of software components based on user-defined requirements. The classification scheme utilizes a dedicated genetic algorithm which evolves a small number of classifiers by dividing the set of available components stored in a database into certain subsets (clusters). Each classifier thus becomes the leader-representative of its cluster. When a user wishes to trace a component he/she identifies the desired characteristics (component profile) which are then compared with the characteristics of the available classifiers. The closest classifier matching the required characteristics over a user-defined threshold will result in the "winning" set of components belonging to its cluster, which are presented to the user in descending matching fitness. We have validated our methodology over a synthetic dataset of components and the results obtained were very encouraging. Last, we present the web application developed to support the proposed intelligent classification method. © 2006 IEEE.