The paper "Compositional API Recommendation for Library-Oriented Code Generation" addresses the challenge of generating library-oriented code using large language models (LLMs). The authors propose CAPIR (Compositional API Recommendation), a novel approach that decomposes coarse-grained requirements into fine-grained subtasks and retrieves relevant APIs from documentation. CAPIR consists of three main components: a Decomposer to break down tasks, a Retriever to identify APIs, and a Reranker to filter and refine the API recommendations. The effectiveness of CAPIR is evaluated on two benchmarks, RAPID and LOGG, demonstrating significant improvements over existing baselines in API recommendation and library-oriented code generation. The paper also includes a qualitative analysis and practical application validation, showing that CAPIR can accurately recommend APIs for complex tasks and improve code generation performance.The paper "Compositional API Recommendation for Library-Oriented Code Generation" addresses the challenge of generating library-oriented code using large language models (LLMs). The authors propose CAPIR (Compositional API Recommendation), a novel approach that decomposes coarse-grained requirements into fine-grained subtasks and retrieves relevant APIs from documentation. CAPIR consists of three main components: a Decomposer to break down tasks, a Retriever to identify APIs, and a Reranker to filter and refine the API recommendations. The effectiveness of CAPIR is evaluated on two benchmarks, RAPID and LOGG, demonstrating significant improvements over existing baselines in API recommendation and library-oriented code generation. The paper also includes a qualitative analysis and practical application validation, showing that CAPIR can accurately recommend APIs for complex tasks and improve code generation performance.