Deep code comment generation

• DOI: 10.1145/3196321.3196334
• Publication Date: 2018-07-19T13:05:12Z
• Authors: Xing Hu; Ge Li; Xin Xia; David Lo; Zhi Jin
• Keywords: Code (set theory); KPI-driven code analysis; Margin (machine learning)

During software maintenance, code comments help developers comprehend programs and reduce additional time spent on reading navigating source code. Unfortunately, these are often mismatched, missing or outdated in the projects. Developers have to infer functionality from This paper proposes a new approach named DeepCom automatically generate for Java methods. The generated aim understand of applies Natural Language Processing (NLP) techniques learn large corpus generates learned features. We use deep neural network that analyzes structural information methods better generation. conduct experiments large-scale built 9,714 open projects GitHub. evaluate experimental results machine translation metric. Experimental demonstrate our method outperforms state-of-the-art by substantial margin.