It is widely understood that developers’ language and bug reporters’ language are different, and the differences limit traceability between commit logs and bug reports. However, few studies have uncovered the differences and tried to overcome the challenges they present. This paper investigates and deals with these issues. First, I clarify the textual difference and lexical relations between bug reports and commit logs by projecting words into context space with deep learning techniques. I also clarify some limitations of the conventional textual similarity measures between bug reports and commit logs on VSM owing to the textual differences. Second, I propose a novel approach, DeepLink, which automatically analyzes textual information and precisely recovers traceability between...[ Read more ]

It is widely understood that developers’ language and bug reporters’ language are different, and the differences limit traceability between commit logs and bug reports. However, few studies have uncovered the differences and tried to overcome the challenges they present. This paper investigates and deals with these issues. First, I clarify the textual difference and lexical relations between bug reports and commit logs by projecting words into context space with deep learning techniques. I also clarify some limitations of the conventional textual similarity measures between bug reports and commit logs on VSM owing to the textual differences. Second, I propose a novel approach, DeepLink, which automatically analyzes textual information and precisely recovers traceability between commit logs and their corresponding bug reports. Lastly, I evaluate the performance of DeepLink on 10 large open-source projects. The experimental results show that DeepLink outperforms conventional techniques by 17% in F-score on average.