Comment
Author: Admin | 2025-04-28
In line with observations made during the interim evaluation. Beside the size of the knowledge base, the domain specificity of the knowledge plays an important role in the requirement mining process (Casamayor et al., 2010). Generally, we can expect a higher degree of domain specificity to deliver better mining results (Lemaigre et al., 2008) by, for example, including domain-specific requirements (such as “conductor” or “attendant”) beside domain-independent ones (like “manager” or “worker”). As we depict in Section 4.2, we propose two sources of knowledge to fill the knowledge base: in addition to manually imported knowledge, which is commonly used in existing RMSs (Kiyavitskaya & Zannone, 2008; Vlas & Robinson, 2012), the content of the knowledge base can be extended with automatically retrieved knowledge originating from previously processed NLRRs. As we describe when conceptualizing DP2, this should increase the size and domain specificity of the knowledge base. Further, encouraged by the interim evaluation’s findings, we hypothesize that: H2: In a fixed time period, using a RMS that supports semi-automatic requirement mining with imported and retrieved knowledge will result in higher recall than using a RMS that only supports semi-automatic requirement mining with imported knowledge. Because both recall and precision determine requirements quality, any impact on precision also has to be evaluated. However, in automated requirement mining from NLRR, recall is significantly more important than precision because it is a much simpler activity for a requirements engineer to evaluate a set of candidate requirements and reject the unwanted ones than it is to browse through an entire document looking for overlooked ones (Cleland-Huang et al., 2007). Berry et al. (2012) make the same argument by stating that requirement engineering tools that treat NLRR “should be tuned to favor recall over precision because errors of commission are generally easier to correct than errors of omission” (Berry et al., 2012, p. 213). Consequently, the artifact’s design principles primarily address an improvement in the recall rate and do not target precision improvements. Journal of the Association for Information Systems Vol. 16, Issue 9, pp. 799-837, September 2015 822 Meth et al. / Designing RMS Moreover,
Add Comment