The study introduces CAPIR, a Compositional API Recommendation approach, to address challenges in generating library-oriented code. By decomposing tasks, retrieving APIs based on subtasks, and reranking recommendations, CAPIR outperforms existing methods in both API sequence recommendation and code generation tasks.
Large language models have shown impressive performance in code generation but struggle with library-oriented code. CAPIR aims to bridge this gap by recommending APIs based on task decomposition.
The study presents experimental results on benchmarks like RAPID and LOCG to validate the effectiveness of CAPIR. Results show significant improvements in recall@k and precision@k metrics across various datasets.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Zexiong Ma,S... at arxiv.org 03-01-2024
https://arxiv.org/pdf/2402.19431.pdfDeeper Inquiries