Access Type

Open Access Thesis

Date of Award

January 2013

Degree Type

Thesis

Degree Name

M.S.

Department

Computer Science

First Advisor

Andrian Marcus

Abstract

The use of text retrieval techniques on concept location and bug localization yields remarkable benefits. The artifacts found in source code and bug reports contain important information related to the bug localization process. When locating the bugs, it is a programmer's task to formulate effective queries such that most of the predicted terms in the query appear in the relevant defect code, but not in most of the non-relevant source files. These queries are built based on the textual content found in the bug reports, especially the bug title and the description. A large body of research uses bug descriptions to evaluate bug localization techniques using text retrieval. All these studies are conducted under the implicit assumption that the bug description and the relevant source code files share important terms. This paper presents an empirical study that explores this conjecture. We found that bug reports share more terms with the patched classes than with the other classes in the software system. Moreover, the study revealed that the class names are more likely to share terms with the bug descriptions than other code locations. We also found that more verbose parts of the source code, such as, comments share more words. Furthermore, we discovered that the shared terms may be better predictors for bug localization than some other text retrieval techniques, such as, LSI.

Share

COinS