home · mobile · calendar · defenses · 2004-2005 · 

Thesis Defense - Faisal

Toward Automating the Discovery of Traceability Links
Maha Faisal
Computer Science PhD Candidate

Maintaining consistency between a system's software artifacts is a hard characteristic to achieve, when you consider the amount of change software artifacts undergo throughout a system's evolution. Requirements traceability is a technique that when properly applied and managed is a step toward the development of software that meets this characteristic. However, requirements traceability faces many challenges that need to be addressed before it becomes a practical "day to day" software development task, including the challenge of reducing its manual, laborious nature. The need for automated solutions to the requirements traceability problem is also underscored by the fact that traceability plays an important role in various areas of software engineering practice.

Automated solutions to software traceability face a difficult challenge due to the need to handle the numerous software artifacts of multiple types generated during a software life cycle, as well as the large number of relationships that exist between them. Furthermore, these artifacts are typically written using natural language, making them difficult to process programmatically. Additionally, the task of searching these documents manually looking for implicit relationships is time consuming and labor intensive. This situation raises the need to automate the discovery of traceability relationships among various types of software artifacts to make the task of software traceability more feasible and cost effective.

Finding or discovering traceability relationships is the essence of the traceability problem. Once found, these relationships or "links" play important roles in various aspects of software evolution. My research is in the area of software traceability, focusing on the issue of automatically detecting the existence of traceability links between requirements documents, design documents, and source code using machine learning. I have also conducted a study to assess my approach's performance. I was interested in how well humans perform the task of finding relationships compared to finding such relationships automatically.

Committee: Kenneth Anderson, Associate Professor (Chair)
Douglas Sicker, Assistant Professor
Dennis Heimbigner, Research Associate Professor
Gregory Grudic, Assistant Professor
Robert France, Colorado State University
Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:20)