A Systematic Mapping Review: Tracking the Relationships Between Software Artifacts using NLP

In software development, traceability from requirements to realization is essential, yet manual tracing is labour-intensive and prone to errors. Requirements Traceability (RT) is crucial for effective management and impact assessment. This paper explores the application of Natural Language Processing (NLP) in requirements traceability (RT) through a systematic mapping review of literature from 2019 to 2023. Out of 209 initial studies, we selected 49 using stringent criteria. RT approaches were categorised into ontology-based and embedding techniques. Embedding techniques have gained prominence for their ability to capture relationships between artifacts. We identified two primary paradigms in RT: rule-based methods, which use predefined heuristics, and machine learning approaches, including traditional classifiers. Machine learning models significantly improve accuracy and adaptability, especially when paired with advanced embeddings. A notable trend is the increasing reliance on standard datasets, like the CoEST repository, to validate methods, enhance reproducibility, and enable robust comparisons. Despite advancements, challenges persist. Non-functional requirements remain underexplored, and the lack of comprehensive benchmarks limits the generalizability of current approaches. Future research should focus on creating inclusive datasets with diverse requirements and integrating hybrid methods to improve performance. Overall, the study underscores the critical role of embedding techniques in RT while highlighting gaps and opportunities for advancing the field.