google-research-datasets/cfq
收藏资源简介：
Compositional Freebase Questions (CFQ) 是一个专门设计用于测量组合泛化能力的数据集。它包含大量的自然语言问题及其对应的SPARQL查询，适用于语义解析任务。数据集为单语种（英语），包含多个配置和分割，每个配置和分割都有详细的下载和生成数据集大小信息。
Compositional Freebase Questions (CFQ) is a dataset specifically designed to measure compositional generalization capabilities. It contains a substantial number of natural language questions paired with their corresponding SPARQL queries, suitable for semantic parsing tasks. This dataset is monolingual (English), and includes multiple configurations and data splits, with detailed information regarding download access and dataset size for each configuration and split.
数据集概述
数据集描述
- 数据集名称: Compositional Freebase Questions (CFQ)
- 数据集摘要: CFQ 是一个专门设计来衡量组合泛化能力的自然语言问答数据集。每个问题都对应一个针对 Freebase 知识库的 SPARQL 查询，因此也可用于语义解析。
- 支持的任务和排行榜:
- 问答
- 其他
- 语言: 英语 (
en) - 许可证: CC-BY-4.0
数据集结构
数据实例
mcd1
- 下载的数据文件大小: 267.60 MB
- 生成的数据集大小: 42.90 MB
- 总磁盘使用量: 310.49 MB
训练集示例: json { "query": "SELECT count(*) WHERE { ?x0 a ns:people.person . ?x0 ns:influence.influence_node.influenced M1 . ?x0 ns:influence.influence_node.influenced M2 . ?x0 ns:people.person.spouse_s/ns:people.marriage.spouse|ns:fictional_universe.fictional_character.married_to/ns:fictional_universe.marriage_of_fictional_characters.spouses ?x1 . ?x1 a ns:film.cinematographer . FILTER ( ?x0 != ?x1 ) }", "question": "Did a person marry a cinematographer , influence M1 , and influence M2" }
mcd2
- 下载的数据文件大小: 267.60 MB
- 生成的数据集大小: 44.77 MB
- 总磁盘使用量: 312.38 MB
训练集示例: json { "query": "SELECT count(*) WHERE { ?x0 ns:people.person.parents|ns:fictional_universe.fictional_character.parents|ns:organization.organization.parent/ns:organization.organization_relationship.parent ?x1 . ?x1 a ns:people.person . M1 ns:business.employer.employees/ns:business.employment_tenure.person ?x0 . M1 ns:business.employer.employees/ns:business.employment_tenure.person M2 . M1 ns:business.employer.employees/ns:business.employment_tenure.person M3 . M1 ns:business.employer.employees/ns:business.employment_tenure.person M4 . M5 ns:business.employer.employees/ns:business.employment_tenure.person ?x0 . M5 ns:business.employer.employees/ns:business.employment_tenure.person M2 . M5 ns:business.employer.employees/ns:business.employment_tenure.person M3 . M5 ns:business.employer.employees/ns:business.employment_tenure.person M4 }", "question": "Did M1 and M5 employ M2 , M3 , and M4 and employ a person s child" }
mcd3
- 下载的数据文件大小: 267.60 MB
- 生成的数据集大小: 43.60 MB
- 总磁盘使用量: 311.20 MB
训练集示例: json { "query": "SELECT /producer M0 . /director M0 . ", "question": "Who produced and directed M0?" }
数据字段
所有分割和配置中的数据字段相同:
question: 字符串类型query: 字符串类型
数据分割
| 名称 | 训练集 | 测试集 |
|---|---|---|
| mcd1 | 95743 | 11968 |
| mcd2 | 95743 | 11968 |
| mcd3 | 95743 | 11968 |
| query_complexity_split | 100654 | 9512 |
| query_pattern_split | 94600 | 12589 |
| question_complexity_split | 98999 | 10340 |
| question_pattern_split | 95654 | 11909 |
| random_split | 95744 | 11967 |
