We present a recommender system that models and recommends product features for a given domain. Our approach mines product descriptions from publicly available online specifications, utilizes text mining and a novel incremental diffusive clustering algorithm to discover domain-specific features, generates a probabilistic feature model that represents commonalities, variants, and cross-category features, and then uses association rule mining and the k-Nearest-Neighbor machine learning strategy to generate product specific feature recommendations. Our recommender system supports the relatively labor-intensive task of domain analysis, potentially increasing opportunities for re-use, reducing time-to-market, and delivering more competitive software products. The approach is empirically validated against 20 different product categories using thousands of product descriptions mined from a repository of free software applications.

Regulatory standards, designed to protect the safety, security, and privacy of the public, govern numerous areas of software intensive systems. Project personnel must therefore demonstrate that an as-built system meets all relevant regulatory codes. Current methods for demonstrating compliance rely either on after-the-fact audits, which can lead to significant refactoring when regulations are not met, or else require analysts to construct and use traceability matrices to demonstrate compliance. Manual tracing can be prohibitively time-consuming; however automated trace retrieval methods are not very effective due to the vocabulary mismatches that often occur between regulatory codes and product level requirements. This paper introduces and evaluates two machine-learning methods, designed to improve the quality of traces generated between regulatory codes and product level requirements. The first approach uses manually created traceability matrices to train a trace classifier, while the second approach uses web-mining techniques to reconstruct the original trace query. The techniques were evaluated against security regulations from the USA government's Health Insurance Privacy and Portability Act (HIPAA) traced against ten healthcare related requirements specifications. Results demonstrated improvements for the subset of HIPAA regulations that exhibited high fan-out behavior across the requirements datasets.

Automated trace retrieval methods can significantly reduce the cost and effort needed to create and maintain requirements traces. However, the set of generated traces is generally quite imprecise and must be manually evaluated by analysts. In applied settings when the retrieval algorithm is unable to find the relevant links for a given query, a human user can improve the trace results by manually adding additional search terms and filtering out unhelpful ones. However, the effectiveness of this approach is largely dependent upon the knowledge of the user. In this paper we present an automated technique for replacing the original query with a new set of query terms. These query terms are learned through seeding a web-based search with the original query and then processing the results to identify a set of domain-specific terms. The query-mining algorithm was evaluated and fine-tuned using security regulations from the USA government's Health Insurance Privacy and Portability Act (HIPAA) traced against ten healthcare related requirements specifications.