Knowledge Graph Construction for AI-Augmented Code Completion in Software Engineering

Abstract

The rapid evolution of software development has led to an increasing reliance on AIdriven tools for code completion, refactoring, and optimization. Traditional machine learningbased code completion systems often struggle with contextual understanding and adaptability. This paper explores the construction of a Knowledge Graph (KG) for AI-augmented code completion in software engineering, leveraging structured and unstructured code repositories, API documentation, and developer interactions. The proposed KG framework captures semantic relationships between programming constructs, coding patterns, and best practices, facilitating enhanced contextual awareness for AI models. By integrating graph neural networks (GNNs) and transformer-based models, the KG-driven approach improves code suggestion accuracy, reduces redundant completions, and enhances developer productivity. Experimental evaluations on open-source repositories demonstrate significant improvements in code recommendation relevance and efficiency compared to conventional deep learning-based models. This research provides a foundation for intelligent software engineering assistants, paving the way for more robust and context-aware AI-driven development environments.