Keyvan Nayyeri

In early May 2010, Dr. Shonle had a trip to Cape Town, South Africa, to attend one of the most notable conferences on Software Engineering, ICSE 2010. He brought some hot topics as souvenirs for us and we decided to present and discover some of the papers in more details every week. Today I had the first presentation about a paper published on the conference entitled Software Traceability with Topic Modeling written by a research group at the University of California, Irvine.

In essence, Software Traceability is a part of Software Engineering that enables the retrieval of information about project artifacts. There have been many researches in this field to improve the performance and accuracy of Software Traceability specifically in regards to the semantic nature of data. There have been two main approaches to this: retrospective and prospective traceability. The former approach is based on capturing the past data and prospective traceability is based on capturing the current data.

This paper tries to combine prospective Software Traceability approach with Machine Learning approaches, so each of these two can fill in the gap with another to improve the results.

I started my presentation by developing the problem and the motivation to introduce this new technique in this paper then gave some general background information as an introduction. After this, I discussed retrospective and prospective Software Traceability approaches along with their characteristics and some insights for effective traceability.

In the next step I introduced Bayesian learning in a high level and related that to some techniques in Topic Modeling in Machine Learning especially Latent Dirichlet Allocation (LDA). In this paper an efficient algorithm called CVB0 is used to implement Topic Modeling.

Having these background information, the next part of my presentation focused on the new technique that combines prospective Software Traceability with Topic Modeling and its benefits.

This approach is implemented as three tools called Topically-Rich Artifact Search Engine (TRASE), Architecture-Centric Traceability for Stakeholders (ACTS), and Topic-Enhanced Architecture Visualization Tool (TEAM) that were the next topics of my presentation.

TRASE, ACTS, and TEAM are applied to a real world project of moderately large scale developed at University of California, Irvine called ArchStudio which is an Open Source Software and Systems Architecture tool based on Eclipse development platform. I analyzed the timing and accuracy of the results of this experiment reflected in the paper as the last part of my presentation.

I have published the slides from my talk and they’re available for download.