danyarm/ckg-benchmark
收藏资源简介:
CKG Benchmark是一个用于知识检索架构基准测试的数据集,特别关注预结构化知识图在多个领域中的表现。数据集包含52个领域的结构化有向无环图(DAG)、7,928个基准查询以及每个系统的JSONL结果和摘要CSV。数据集强调了预结构化知识图在47个基准领域中的优越性,包括更高的F1分数和更低的令牌成本。此外,数据集还引入了新颖的度量标准,如检索密度分数(RDS)、跳深度F1和每正确答案成本(CPCA)。数据集内容涵盖教育和技术等多个领域,旨在为知识检索和表示提供全面的基准测试工具。
The CKG Benchmark is a dataset designed for benchmarking knowledge retrieval architectures, with a particular focus on the performance of pre-structured knowledge graphs across multiple domains. The dataset includes structured directed acyclic graphs (DAGs) for 52 domains, 7,928 benchmark queries, and JSONL results along with summary CSVs for each system. It highlights the superiority of pre-structured knowledge graphs in 47 benchmarked domains, demonstrating higher F1 scores and lower token costs. Additionally, the dataset introduces novel metrics such as Retrieval Density Score (RDS), Hop-Depth F1, and Cost Per Correct Answer (CPCA). Covering various fields including education and technology, the dataset serves as a comprehensive tool for benchmarking knowledge retrieval and representation.




