Neural Network-based Approach for Source Code Classification to Enhance Software Maintainability and Reusability

Abstract

One of the most essential resources in software development is a program's source code. When a developer starts coding from scratch for each project, it takes more time and money to develop. When it comes to software reuse and maintainability, topic modelling is critical since it may be difficult for developers to remove outdated source code from huge systems with a lot of code. In a variety of ways, topic modelling approaches have been used to analyze and model source codes. Using various statistical techniques and methodologies, several pieces of research have been done to extract topics from source codes. Attempts to extract topics from method names, identifiers, and comments are the most common. These topic extraction approaches are interdependent, and if software best practices aren't followed in older systems, it might cause chaos. Motivated by these observations, in this paper, the authors have conducted a study on extracting source code using a JavaParser and predicting the source code functionality name through the artificial neural network model. It shows an average accuracy of 88 percent semantic function prediction rate. This is a new approach for topic modelling and the first attempt in building an artificially intelligent model to predict the semantic function name of the source code.