Vector Representation of Abstract Program Tree for Assessing Algorithm Variety for the Same Purpose

摘要

There are various ways to realize programs to meet the same requirements. Therefore, for evaluation of a program code, it is necessary not only to satisfy the requirement for it but also to evaluate how it achieves the purpose. We think that the knowledge of such alternative ways to achieve the same purpose is important for program education and software engineering. In this paper, we propose a method to analyze how the program achieves requirements. We propose particularly a vector representation that appropriately indicates the structure of an abstract syntax tree. We confirmed that our proposal not only analyzes fundamental programs as appears in textbooks effectively but also classifies algorithms that are adopted in various programs submitted to a programming contest. Based on this confirmation, we further investigated the relationship between the class name of the Java language and the program structure. As a result, it was shown that classes with similar program structure are named by certain similar linguistic expressions. Therefore, we conclude that our proposed method is a useful basis for representing the diversity of programs on vector space.