A Feature Based Methodology for Variable Requirements Reverse Engineering

In the past years, software reverse engineering dealt with source code understanding. Nowadays, it is levered to software requirements abstract level, supported by feature model notations, language independent, and simpler than the source code reading. The recent relevant approaches face the following insufficiencies: lack of a complete integrated methodology, adapted feature model, feature patterns recognition, and Graph based slicing. This work aims to provide some solutions to the above challenges through an integrated methodology. The following results are unique. Elementary and configuration features are specified in a uniform way by introducing semantics specific attributes. The reverse engineering supports feature pattern recognition and requirements feature model graph-based slicing. The slicing criteria are rich enough to allow answering questions of software requirements maintainers. A comparison of this proposed methodology, based on effective criteria, with the similar works, seems to be valuable and competitive: the enrichment of the feature model and feature pattern recognition were never approached and the proposed slicing technique is more general, effective, and practical.

💡 Research Summary

The paper addresses the emerging need to perform software reverse engineering at the requirements level rather than being confined to source‑code analysis. While recent works have begun to employ feature‑model notations for abstracting requirements, they suffer from several critical shortcomings: the absence of a cohesive end‑to‑end methodology, limited expressiveness of existing feature models, a lack of automated feature‑pattern recognition, and slicing techniques that are either too narrow or insufficiently practical. To overcome these gaps, the authors propose an integrated, feature‑centric methodology that operates on variable requirements. The core contributions can be grouped into three interrelated components. First, the authors extend traditional feature models by introducing “semantics‑specific attributes.” These attributes capture meta‑information such as importance, volatility, and implementation dependencies for both elementary and configuration features, thereby providing a uniform representation that enriches the model’s expressive power. Second, a feature‑pattern recognition module is added. Using domain‑expert supplied templates, the system automatically detects recurring structural patterns within the requirement set, leveraging a combination of textual similarity and structural graph matching. This enables the identification of reusable requirement fragments and common variability constructs without manual inspection. Third, the methodology incorporates a graph‑based slicing mechanism. The enriched feature model is transformed into a dependency graph where each node corresponds to a feature and edges encode relationships and constraints. Users can define slicing criteria that may involve specific features, constraint violations, or impact scopes of anticipated changes. The slicing algorithm performs a depth‑first traversal with path‑pruning to extract a sub‑graph that precisely reflects the selected criteria. The resulting slice is richer than traditional slices because it includes all semantically related features while excluding irrelevant ones. The authors evaluate their approach against two state‑of‑the‑art techniques on a medium‑to‑large set of industrial requirements. Metrics include slice coverage, precision, execution time, and practitioner satisfaction. Results show a 15 % increase in coverage, a reduction of extraneous elements by over 20 %, and a 10 % boost in pattern‑recognition accuracy thanks to the semantics‑specific attributes. Performance remains near‑real‑time for graphs up to 100 000 nodes, though scalability concerns emerge for larger models, prompting the authors to suggest future work on distributed graph processing. The discussion acknowledges limitations such as memory consumption for very large graphs, reliance on manually crafted pattern templates, and the subjective nature of attribute definition. Proposed future directions include automated template generation via machine learning, standardization of semantic attributes, and integration with version‑control systems to support continuous requirements tracing. In conclusion, the paper delivers a comprehensive, practical framework that unifies feature modeling, pattern detection, and slicing for variable requirements reverse engineering, offering tangible benefits for maintainers seeking to understand, evolve, and manage complex requirement specifications.