Ambiguity in software requirements can often lead to misinterpretations, which may hinder the proper implementation of systems to meet user needs. Software requirements are fundamentally stated in natural language, making them subject to various interpretations. Thus, effective understanding of these requirements is crucial for all stakeholders involved.
Requirements encompass what should be developed and serve as a vital guide through the software development process. They embody both functional and non-functional aspects of projects. However, the natural language used to formulate these requirements can be inconsistent, incomplete, and ambiguous. Such inconsistencies highlight the necessity for precise identification of requirement entities, which are central to understanding the core concepts within the documentation.
Traditionally, manually extracting entities from requirement statements is labor-intensive and can lead to issues such as duplication or omission. Existing manual techniques, often reliant on human judgment, exacerbate the risk of errors in identifying co-referent entities and resolving ambiguities. Manual approaches are not only time-consuming; they fall short in achieving the level of consistency required for effective software requirement analysis.
To address these challenges, automated entity extraction methods have arisen, notably through the use of parsing techniques. The ongoing evolution of technology has influenced the development of these methods, although most existing solutions leverage machine learning or deep learning models, which necessitate extensive pre-training on domain-specific data.
This study introduces an innovative hybrid approach that integrates two widely recognized parsing techniques: constituency parsing and dependency parsing. Both techniques are employed to extract noun phrase-based entities from software requirements. Importantly, the proposed approach circumvents the need for extensive pre-training datasets, instead utilizing pre-trained parsers grounded in general linguistic corpora. This methodology reduces the need for laborious manual input while enhancing cross-domain applicability.
The dataset for this study was manually curated, consisting of software requirements from the healthcare domain. It was annotated to establish a reliable gold standard for entity extraction evaluation. The research compares outputs from individual constituency and dependency parsing techniques against those from the newly developed hybrid approach, which combines results post-duplication removal to improve noun phrase span coverage.
The experimental results demonstrated that the proposed hybrid parsing method achieved the highest performance metrics among evaluated methods. While the constituency parser, on its own, yielded the most precise extractions, it fell short in recall. Conversely, the hybrid approach excelled, achieving commendable recall and F1-score metrics while maintaining precision levels akin to those recorded for the constituency parser. Notably, the hybrid method showcases improved extraction efficacy by leveraging grammatical structures inherent to the requirements statements without necessitating extensive domain-specific training configurations.
The findings underscore the potential of the hybrid parsing approach to advance the state of the art in requirements entity extraction. As effective entity identification is fundamental to various tasks—including glossary creation, feature extraction, and project size estimation—this research paves the way for streamlined processes in software development, ultimately fostering greater clarity and understanding amongst stakeholders. Through the advancements detailed here, the realm of software requirements stands to benefit from methodologies that prioritize efficiency and accuracy, mitigating the risks posed by the inherent ambiguities of natural language.
The subsequent sections of this research delve deeper into literature reviews on existing approaches, offer background information on the parsing techniques used, and detail the methodology and evaluation metrics employed throughout the study. The implications and recommendations drawn from this research could redefine best practices in requirements analysis, marking a pivotal shift towards more efficient systems development.