DocScribe: A Transformer-based Approach for Generating Accurate and Fluent Documentation References.

Abstract

In the ever-evolving landscape of software development, code documentation plays a pivotal role. Clear, concise, and well-structured documentation not only aids developers in understanding the codebase but also contributes to maintainability, efficiency, and overall software quality. However, manually writing documentation comments can be time-consuming and error-prone. In this research project, the author explores the realm of natural language processing (NLP) to develop ‘DocScribe’, an innovative solution for generating documentation comments (JSDoc references) directly from programming source code. By analyzing code structure and keyword usage, DocScribe can infer the functionality and purpose of code segments. This allows the tool to generate more comprehensive documentation comments that explain not just what the code does using a single sentence. In this research project, the author leverages state-of-the-art transformer architectures to develop the proposed NLP-based tool that dissects code snippets, identifies relevant entities (such as functions, classes, and variables), and crafts contextually appropriate documentation comments. This research aims to streamline the documentation process, empower developers, and elevate quality software engineering practices by bridging the gap between code development and in-code documentation.