Extraction of editing processes for modifying a program to another purpose based on representation learning
Wakana Hashimoto, Yo Amano, Yutaka Takesue, Kazuhiro Takeuchi · 2021
The purpose of this paper is to clarify the editing process in creating computer programs. We do not record those processes of many people, but propose a way to find the process of editing a program from a set of submitted programs for a wide range of program tasks. In this way, we would like to clarify how to edit and modify a simple program to another more complex program. In the proposed method, we find the programmers' editing process by training a representation learning model that converts the tree structure into a vector representation, and clustering with the distance between programs that are embedded in a vector space. As a result, we show a transition diagram of the editing process we extract from various submitted programs.