Bugzie Model Overview
In Bugzie, the problem of automatic bug triaging is modeled as follows:
given a bug report, find the developer(s) with the most fixing
capability/expertise with respect to the reported technical issue(s).
Existing approaches view this problem as a classification problem: each
developer is considered as a class for bug reports in which their
characteristics are learned via his/her past fixed reports. An unfixed bug
report will be assigned to the developer(s) corresponding to the most
relevant/similar class(es) to the report.
In contrast, Bugzie considers this as a ranking problem: for each given
bug report, it determines a ranked list of developers who are most capable of
handling the reported technical issue(s). Thus, instead of learning the
characteristics of each class/developer based on his past fixed reports, Bugzie
determines and ranks the fixing capability/expertise of the developers toward
the technical aspects by modeling the correlation/association of a developer and
a technical aspect. That is, if a developer has higher fixing correlation with a
technical aspect, (s)he is considered to have higher capability/ expertise on
that aspect, and (s)he will be ranked higher.
Because “technical aspect” is an abstract concept, with potential different
levels of granularity, Bugzie models them via their corresponding descriptive
technical terms. That is, a technical aspect is considered as a collection of
technical terms that are extracted directly from the software artifacts in a
project, and more specifically from its bug reports. Bugzie utilizes the fuzzy
set theory to model the fixing correlation/association between developers and
the technical terms/aspects, which is used to recommend the most capable fixers
for a given bug report. Bugzie also uses the locality of fixing activity to
select the fixer candidates, and uses the levels of correlation between the
fixers and terms to identify the most correlated/important terms for each
developer.
