Knowledge Graphs for Software Engineering: A Systematic Review of Applications, Techniques, and Emerging LLM Integration

Abstract

Knowledge graphs (KGs) have proven to be a powerful knowledge representation and reasoning paradigm for automating software engineering (SE) tasks such as bug localization, API recommendation, and vulnerability assessment. Although the topic is gaining traction, an overview of the use of KGs in SE is lacking. In this paper, we conduct a systematic review of 60 papers (2016-2025) on the application of KGs in SE. We pose four research questions, examining application areas, KG construction and reasoning techniques, data and tools, and future trends and challenges.

We find that KGs are primarily used for code understanding and bug detection (40%), API development (25%) and security/vulnerability analysis (20%). Construction techniques encompass ontology design, information extraction, graph embeddings and neural KG completion. Although numerous papers are able to demonstrate high performance in controlled experiments (e.g., >80% precision), issues around scalability, updateability and deployment remain. Recent studies focus on the use of KGs in conjunction with large language models for vulnerability and code analysis. We introduce a classification of applications, review current techniques, and discuss possible directions for future work, including the need for benchmarks, ethical use cases and better integration with LLMs.