Mining Requirements Links

[Context & motivation] Obtaining traceability among requirements and between requirements and other artifacts is an extremely important activity in practice, an interesting area for theoretical study, and a major hurdle in common industrial experience. Substantial effort is spent on establishing such links, and keeping them up to date, in any large project – even more so when requirements refer to several generations of a product, or to a product family. [Question/problem] While most research is concerned with ways to reduce the effort needed to establish and maintain traceability links, a different question can also be asked: how is it possible to harness the vast amount of implicit (and tacit) knowledge embedded in already-established links? Is there something to be learned about a specific problem or domain, or about the humans who establish traces, by studying such traces? [Principal ideas/results] In this paper, we present preliminary results from a study applying different machine learning techniques to an industrial case study, together with an assessment of how what is learned from pre-existing traces can affect further product development. [Contribution] Reshaping traceability data into knowledge can contribute to more effective automatic tools to suggest candidates for linking, and at the same time provide some insight into both the domain of interest (e.g., how different writing style and vocabulary is used in marketing requirements vs. technical requirements) and about the actual implementation techniques (e.g., how specific user requirements are refined into a technical specification).