Another nail to the coffin of faceted controlled-vocabulary component classification and retrieval

Abstract

Our research centers around exploring methodologies for developing reusable software, and developing methods and toofs for building with reusable software. In this paper, we focus on reusable software component retrieval methods that were developed and teated in the context of ClassServer, an experimental library tool developed at the University of Qu6bec at Montr6al to explore fssues in software reuse [15]. The methods dfscusaed in tbfa paper fall into two categori~1) string search-based retrieval metbod$ and 2) keyword-based retrieval methods. Both kinds of methods have been implemented and tested by researchers, both in the context of software repositories (see e.g. [6,9]) and in the context of more traditional document tibrarles (see e.g. [2,25]). Experiments have shown that keyword-based methods, which require some manual, laborintensive pre-proceashrg, performed only marginally better than the entfrely mechanical strhtgsearch methods (see e.g.[6, 2S]), raising the issue of cost-effectivene= of keyword-based methods as compared to string search baaed methods. In this paper, we describe an implementation and experiments which attempt to brfng the two khtds of methods to a level-playing field by: 1) automating as much of the pre-processing involved hr controlled vocabulary-based methods as possible to address the crds issue, and 2) using a realistic experimental setting in which queries consist of problem statemenlx rather than component specifications, in whjch query results are aggregated over several trials, and in which recaU measures take into account overlapping components. Our experiments showed that string search based methods performed better than semi-controlled vocabulary-based method$ which goes further in the direction of more recent component retrfeval experiments which challenged the superiority of controlled vocabulary based clarification and retrieval of components (see e.g. [61).