Abstract:
Requirement engineering is the most crucial phase of the software development lifecycle as it determines the need and the expectations of the stakeholders from the software. The requirements engineering phase therefore plays a key role in determining the success or failure of the software project. Emerging trends in software development such as globalization and large scale development of software have made it difficult to carry out requirement elicitation using the traditional methods such as interview, observation, joint application development et cetera hence the automated methods which facilitates universally dispersed stakeholders to gather requirements. This automated method has however been found to generate requirements faster than can be meaningfully analyzed in a central repository. To address the foregoing problem, this paper uses an empirical prior latent Dirichlet allocation model to carefully break the large volume of requirements gathered in a central repository into manageable sizes. Requirements data for Student Records Management System and Timetable Scheduling Systems were gathered by online elicitation method and the epLDA model was used to classify these datasets into manageable sizes. Using precision, recall and measures, the epLDA model was able to achieve an average accuracy of 85.5%. This model outperforms previously used models namely, Latent Dirichlet allocation model and Probabilistic Latent Semantic Analysis model by 13.0% and 17.88% respectively. Unlike the previously used models, the epLDA model was found to be sensitive and adaptive to its environment. We therefore recommend the epLDA model for use in classification of overwhelming requirements requests that are being gathered by online elicitation into manageable sizes for the purpose of effective analysis.
Keywords: Adaptive, requirements analysis, prior, latent Dirichlet allocation, latent semantic analysis.
