ReqNet and ReqSim: A Network and Semantic Similarity Dataset of Requirements From the Tree Structure of System Requirement Specifications
Chandan Sahu, Rahul Rai, Margaret M. Wiecek, David Gorsich · Journal of Computing and Information Science in Engineering · 2024
Abstract Systems are developed as a solution to the problem space defined by their requirements. The requirements are acquired during the elicitation process. The creative nature of the elicitation process, proprietary nature of requirements, the need of extensive preprocessing, and the diverse techniques for analysis restricts the development of a requirement dataset. There exists no standard method to create a requirement dataset. Thus, we devise a semi-formal method to create a multi-purpose open-source requirement dataset that harnesses human knowledge in the system requirement specification documents (SyRSDs). We devise a method to extract a tree from each SyRSD. Our dataset has three forms. (1) ReqList, a list of requirements from 86 distinct systems with their document structure in pure text form. The 12,701 requirements are ready to leverage natural language processing techniques and unsupervised machine learning techniques; (2) ReqNet, a large network of requirements consisting of 17,375 nodes to deploy graph-theoretic algorithms for requirement engineering. ReqNet portrays small-world network characteristics with an average distance of ≈9.5619 links; (3) ReqSim, a dataset consisting of 10,933 pairs of requirements annotated with their similarity scores. ReqSim enables sentence-level supervised learning tasks to exploit the semantics of requirements. The similarity scores are coherent with human knowledge. The dataset is grounded by the tree structure of SyRSDs. The tree structure resonates with the hierarchical nature of the requirement allocation process. The ReqList, ReqNet and ReqSim dataset will facilitate the deployment of modern computing algorithms for requirements engineering.