Comment Generation by LSTM for Procedure Learning System

Abstract

For beginners of the program, we are developing a system that learns the sequence of program procedures on a tablet computer. We would like to create many exercises for the procedure learning system. Therefore, a system for automatically generating procedures is needed. In this study, in order to generate a procedure from source code, learn source code and corresponding comment using Encoder-Decoder model implemented by LSTM of neural networks and generate comment corresponding to new programs. Also, when targeting learners who are a little familiar with the program, we also show that we can deal with preprocessing comments corresponding to the program.