HSCRD: Hybridized Semantic Approach for Knowledge Centric Requirement Discovery

Abstract

There is a necessity for a requirement recommendation system that can eliminate the old tedious recommendation discovery process. In this paper, a hybrid semantic approach for knowledge-centric requirement discovery has been proposed. The proposed HSCRD framework takes stakeholders’ interactions and preprocesses it. The individual keywords obtained are input into the TF-IDF model to yield the documents from the Requirement Specification Document Repository. The index words of these documents are extracted and are linked with Upper Domain Ontologies. The ontologies are grown by computing the semantic similarity measures, namely, Jaccard similarity and SemantoSim similarity. These grown ontologies are submitted to the Wikidata API, Freebase API, and DBPedia API to yield the Enriched Domain Ontologies. The Enriched Domain Ontologies, as features, are passed into Bi-gram and Tri-gram models. Using Bi-gram and Tri-gram of these features, input is given to the Bagging model for classification. Bagging is chosen with SVM and a highly complex Decision Tree classifier. Finally, the recommended documents and ontologies features are passed individually with respect to the classified documents through TF-IDF and semantic similarity pipeline in order to recommend the individual requirements. The proposed HSCRD achieves the highest average Accuracy with the Precision of 93.18%, Recall of 95.69%, Accuracy of 94.43%, F-Measure of 94.42%, a low FDR of 0.07, and a very high nDCG of 0.96.