Semantic Code Finder: an Efficient Semantic Search Framework for Large-Scale Codebases

We present SEMANTIC CODE FINDER, a framework for semantic code search that delivers high-level search performance and supports multiple programming languages. Leveraging code summaries, it enables meaningful semantic code search by extracting the core content of code methods and using this information for search queries. Evaluated on largescale codebases, SEMANTIC CODE FINDER demonstrates its effectiveness in outperforming existing open-source code search tools, achieving higher recall and precision rates. It delivers superior search performance across Java, Python, and C++. Notably, SEMANTIC CODE FINDER outperforms CodeMatcher, a previously successful semantic code search tool, by approximately 41% in terms of MRR. Moreover, it shows consistent performance across Java, Python, and C++ languages, highlighting its robustness and effectiveness. Currently, it is being used as a code search service for a significant amount of source code within Samsung Electronics, meeting the needs of its developers.