Data science processes often rely on the R programming language for its rigorous statistical methods, yet AI agents frequently struggle to navigate its vast ecosystem. Traditional retrieval methods focus solely on the text-based description of functions, often missing the crucial context of the data itself.

Researchers at The Hong Kong Polytechnic University have introduced DARE (Distribution-Aware Retrieval Embedding), a lightweight model designed to bridge this gap. By fusing function metadata with specific data distribution features, DARE ensures that the AI selects the most appropriate statistical tools for the specific dataset at hand. This distribution-aware approach allows the system to understand not just what a function does, but which data characteristics it is best suited for.

The project also introduces RPKB, a comprehensive knowledge base of over 8,000 packages from the Comprehensive R Archive Network (CRAN). When integrated into a new framework called RCodingAgent, the system achieved a 93.47% accuracy rate in finding the right tools, significantly outperforming larger, general-purpose models. This advancement suggests a future where AI can handle highly specialized academic and professional statistical tasks with much greater reliability.

Data science processes often rely on the R programming language for its rigorous statistical methods, yet AI agents frequently struggle to navigate its vast ecosystem. Traditional retrieval methods focus solely on the text-based description of functions, often missing the crucial context of the data itself.

Researchers at The Hong Kong Polytechnic University have introduced DARE (Distribution-Aware Retrieval Embedding), a lightweight model designed to bridge this gap. By fusing function metadata with specific data distribution features, DARE ensures that the AI selects the most appropriate statistical tools for the specific dataset at hand. This distribution-aware approach allows the system to understand not just what a function does, but which data characteristics it is best suited for.

The project also introduces RPKB, a comprehensive knowledge base of over 8,000 packages from the Comprehensive R Archive Network (CRAN). When integrated into a new framework called RCodingAgent, the system achieved a 93.47% accuracy rate in finding the right tools, significantly outperforming larger, general-purpose models. This advancement suggests a future where AI can handle highly specialized academic and professional statistical tasks with much greater reliability.