[Research] Untangling Composite Changes Using Tree-based Convolution Neural Network

Cong Li, a master’s student of our laboratory, presented our paper in the 2021 March Meeting of IEICE Technical Committees on Software Science, held online.

Authors: Cong Li, Takashi Kobayashi（Tokyo Tech）
Title: Untangling Composite Changes Using Tree-based Convolution Neural Network
Book Title: 信学技報, vol. 120, no. 407, SS2020-46, pp. 108-113, 2021年3月.
Abstract:
Developers often bundle unrelated changes in a single commit, thus creating a so-called composite commit. Composite commit is problematic because it makes code review, reversion, and integration of these commits harder. Recent
researches have attempted to use the information of Abstract Syntax Tree (AST) to untangling composite commits. However, they did not make full use of the AST structure information. To make full use of AST structure information to untangle a
composite commit. First, we predict the relationship between two code fragments using a Tree-based CNN model, which can capture both the structural and lexical information of the code fragment. Second, we cluster these code fragments according to
their relationship. Third, we evaluated whether our approach can untangle composite commits correctly.