Octopus Benchmark
收藏资源简介:
Octopus Benchmark是由康奈尔大学团队构建的面向多表数据发现的评估基准,包含5个支持独立表回答的异构数据集和2个需要表连接操作的复杂场景数据集。该基准创新性地同时提供表级和单元格级标注,支持细粒度检索精度评估,数据来源涵盖真实世界的数据湖和云存储系统。通过轻量化的列头嵌入索引设计,该数据集可高效支持自然语言查询到多表联合检索的端到端验证,主要应用于智能数据发现、NL2SQL优化等数据库前沿领域。
The Octopus Benchmark is an evaluation benchmark for multi-table data discovery, developed by a research team from Cornell University. It includes 5 heterogeneous datasets that support single-table query answering, and 2 complex scenario datasets that require table join operations. Innovatively, this benchmark provides both table-level and cell-level annotations, enabling fine-grained retrieval accuracy evaluation. Its data sources are derived from real-world data lakes and cloud storage systems. With a lightweight column header embedding index design, this benchmark can efficiently support end-to-end validation of natural language queries for multi-table joint retrieval. It is mainly applied in cutting-edge database research fields such as intelligent data discovery and NL2SQL optimization.
Octopus数据集概述
数据集基本信息
- 数据集名称:Octopus
- 核心内容:与论文《Octopus: A Lightweight Entity-Aware System for Multi-Table Data Discovery and Cell-Level Retrieval》相关的源代码。
数据集结构与内容
- 主要组成部分:源代码。
- 关联模块:
- 基准测试生成器:位于
benchmark_generator/目录下。 - 实验:位于
experiments/目录下。
- 基准测试生成器:位于
环境配置要求
- Python版本:3.12.2
- 依赖安装:通过
requirements.txt文件安装所有必需的包。

- 1Octopus: A Lightweight Entity-Aware System for Multi-Table Data Discovery and Cell-Level Retrieval康奈尔大学 · 2026年



