WikiSQL

该数据集名为WikiSQL，包含了超过8万个自然语言问题，这些问题分布在2.4万个表格及其对应的黄金SQL查询中。它专为文本到SQL的任务设计，即从自然语言问题生成SQL查询。该数据集的性能评估采用两种准确度类型：逻辑形式（LF）和执行（EX）。LF准确度检查生成的查询是否与黄金查询完全匹配，而EX准确度则检查预测查询与黄金查询输出的结果是否相等。规模上，该数据集涵盖了超过8万个问题，涉及2.4万个表格。其任务类型为文本到SQL。

The dataset named WikiSQL contains over 80,000 natural language questions paired with 24,000 tables and their corresponding gold-standard SQL queries. It is specifically developed for the text-to-SQL task, which focuses on generating SQL queries from natural language questions. For performance evaluation, two accuracy metrics are employed: Logical Form (LF) accuracy and Execution (EX) accuracy. LF accuracy checks if the generated SQL query exactly matches the gold-standard query, while EX accuracy verifies whether the outputs of the predicted query and the gold-standard query are consistent. In terms of scale, this dataset covers more than 80,000 questions across 24,000 tables, and its target task is text-to-SQL.

搜集汇总

背景与挑战

背景概述

WikiSQL是一个专为文本到SQL任务设计的数据集，包含超过8万个自然语言问题及其对应的黄金SQL查询，覆盖2.4万个表格。该数据集通过逻辑形式（LF）和执行（EX）两种准确度类型进行评估，旨在支持从自然语言生成SQL查询的研究和应用。