RioCC: Efficient and Accurate Class-Level Code Recommendation Based on Deep Code Clone Detection.

Area of Science:

• Software Engineering
• Artificial Intelligence
• Computer Science

Background:

• Current code recommendation methods are limited to local contexts (method/API-level).
• Class-level code recommendation is needed to handle large code spaces and preserve structural information.
• Existing approaches lack efficiency and scalability for large-scale code recommendation.

Purpose of the Study:

• To propose RioCC, a novel class-level code recommendation framework.
• To leverage deep forest-based code clone detection for efficient candidate space reduction.
• To improve recommendation efficiency and accuracy in large-scale code environments.

Main Methods:

• RioCC employs a coarse-to-fine candidate reduction strategy.
• A quick search-based filtering module performs initial candidate screening.
• A deep forest-based analysis with cascade learning and multi-grained scanning refines similarity assessment.

Main Results:

• RioCC outperforms state-of-the-art methods (CCLearner, Oreo, RSharer) on a large dataset (192,000 clone pairs).
• The framework significantly accelerates the recommendation process while maintaining comparable detection accuracy.
• RioCC demonstrates superior performance across four types of code clones.

Conclusions:

• Class-level code recommendation can be effectively modeled as a staged retrieval and refinement problem.
• RioCC offers an efficient and scalable solution for large-scale code recommendation.
• Integrating lightweight filtering with deep forest-based learning is practically valuable.