Text Mining to Concept Mining: Leads Feature Location in Software System

摘要

In the agile application development environment, automatically identifying relevant components in a large complex software system for software maintenance is still remain a research problem with the proliferation of software applications. Earlier, concept mining with formal concept analysis was one of the commonly applied techniques for legacy software systems of small to medium size. Recently, text mining is being widely used for locating features or concerns in a large complex software system. Nevertheless, the literature study reveals that combining text mining with other techniques always yield better accuracy in locating features. Even though it is efficient, applying formal concept analysis on the large systems poses limitation due to its exponential time complexity in constructing concept lattices. In this research work, a model is devised to combine text mining and concept mining for large systems. The unsupervised machine learning technique, Latent Dirichlet Allocation modeling also called as Topic Modeling is used to reduce the feature space on which K-Means clustering is applied to cluster the related documents and formal concept analysis is carried out on individual clusters. Three open source software systems namely JEdit, ArgoUML and JabRef are considered for the experimental study. The empirical evaluation of feature location measure of the proposed model shows a significant improvement in terms of accuracy, scalability, flexibility and efficiency over the contemporary methods existing in the literature.