ExCS: accelerating code search with code expansion.

Area of Science:

• Computer Science
• Software Engineering
• Information Retrieval

Background:

• Developer productivity relies on efficient code search within large codebases.
• Deep learning models offer high accuracy but face computational challenges due to dimensionality and slow retrieval times.
• Traditional Information Retrieval (IR) methods are faster but less accurate than deep learning models.

Purpose of the Study:

• To introduce ExCS, an innovative code search tool designed to accelerate code search without sacrificing accuracy.
• To address the computational challenges posed by high-dimensional deep learning models in code search.
• To improve the efficiency of online code retrieval processes.

Main Methods:

• ExCS utilizes code expansion in an offline phase, predicting potential queries to enrich code semantics.
• During online retrieval, ExCS employs IR-based methods to identify a focused set of relevant code candidates.
• Evaluations were performed on the Java dataset from the CodeSearchNet corpus.

Main Results:

• ExCS achieved a 90% reduction in retrieval duration.
• ExCS maintained an impressive 99% retrieval accuracy.
• The tool effectively balances speed and accuracy in code search.

Conclusions:

• ExCS offers a significant improvement in code search efficiency.
• The developed method enhances semantic depth through offline code expansion.
• ExCS provides a practical solution for fast and accurate code retrieval in large codebases.