A Richly Attributed Dataset for Talent Intelligence Evaluation

Talent evaluation is an important part of tapping outstanding talents, rationally allocating human resources and stimulating innovation and entrepreneurship. However, the development of benchmark datasets suitable for the intelligent evaluation of talent lags behind. Current datasets for talent intelligence evaluation are often collected from diverse sources or integrated by enterprise HRM. This heterogeneous aggregation presents significant challenges for crafting high-quality, fine-grained talent evaluation algorithms. Moreover, existing studies largely overlook factors such as work background, qualifications, and prizes, which significantly influence talent evaluation. To address these issues, we have developed a Richly Attributed Talent (RAT) dataset. This dataset is comprised of authentic, multi-sourced data on professional and technical talent, collected over an extended period by systems such as management platforms for professional and technical personnel and continuing education. Utilizing data governance techniques like fine-grained talent attribute vectorization and differential privacy anonymization, the RAT dataset comprises 244,610 talent samples, each annotated with 38 attributes including prizes, thesis, projects, and work background. To our knowledge, the RAT dataset is the most extensive and detailed available for talent intelligence evaluation, promising significant enhancements in research on professional talent intelligence applications. Furthermore, we validated the validity of the RAT dataset through multiple baseline empirical analyses.