Software cost estimation is a ubiquitous task in software project portfolio management. Analogy-based estimation approaches are particularly suitable to reuse past project experience to create estimate proposals.
However, to find project data of past, similar projects, one usually must rely on similarity measures based on explicit project metrics, like a team´s project experience, the number of interfaces, etc. This requires the consistent collection of such project metrics over time, which can be tedious, error-prone and expensive.
This paper proposes an alternative approach: to use text artifacts arising in the regular course of a project to determine the similarity of project entities. Use cases´ text descriptions are used as entities; a variation of self-organizing maps is used to determine their similarities. The paper points out possible applications and limitations of the approach, analyzes several real-world use case sets, describes problems and suggests rules for writing suitable use cases.