Mybatis-Spider

The Mybatis-Spider dataset is a large-scale, high-quality resource designed for the training and evaluation of models for generating Java MyBatis Mapper XML files from natural language. It is derived from the well-known Spider text-to-SQL dataset through a systematic restructuring and optimization process to better align with real-world software development practices. The dataset addresses the task of generating executable MyBatis Mapper files based on a combination of natural language descriptions, database schemas, and query parameters. To enhance its practical relevance, samples from the Spider dataset with identical SQL logic were consolidated, and queries with similar structures were merged to reflect the use of parameterization in actual MyBatis development. The dataset includes 5,653 pairs of data, each containing a detailed natural language description, parameters, the target Mapper XML file, and the corresponding database context. All Mapper files were generated with the assistance of GPT-4o and have been manually verified to ensure their syntactic correctness and execution accuracy in a live database environment, making it a robust benchmark for code generation tasks in the Java ecosystem.