nlp赛事_关系抽取任务

百度2020年提供了5个赛道
事件抽取任务
关系抽取任务
本文内容
关系抽取就是根据SPO三元组标注训练模型,使得模型能够自动识别SPO三种类别

任务目标是在给定的文本句子中,根据预先定义的schema集合,抽取出所有满足 schema 约束的 SPO 三元组。schema 定义了关系 P 以及其对应的主体 S 和客体 O 的类别,根据 O 类型的复杂程度可以划分为以下两种:

  1. 简单 O 值:也就是说 O 是一个单一的文本。例如,「妻子」关系的 schema 定义为:
{
    S_TYPE: 人物,
    P: 妻子,
    O_TYPE: {
        @value: 人物
	}
}

简单 O 值是最常见关系类型,去年竞赛中所发布的所有 schema 都属于这种类型。为了保持格式统一,简单 O 值类型的 schema 定义通过结构体保存,结构体中只有一个 @value 字段存放真正的 O 值类型。

  1. 复杂 O 值:也就是说 O 是一个结构体,由多个语义明确的文本共同组成,多个文本对应了结构体中的多个槽位(slot)。例如,「饰演」关系中 O 值有两个槽位 @value 和 inWork,分别表示「饰演的角色是什么」以及「在哪部影视作品中发生的饰演关系」,其 schema 定义为:
{
    S_TYPE: 娱乐人物,
    P: 饰演,
    O_TYPE: {
        @value: 角色
        inWork: 影视作品
    }
} 

在复杂 O 值类型的定义中,@value 槽位可以认为是该关系的默认 O 值槽位,对于该关系不可或缺,其他槽位均可缺省。

输入/输出:

输入:schema约束集合及句子sent
输出: 句子sent中包含的符合给定schema约束的三元组知识Triples

数据简介 Dataset

本次竞赛使用的DuIE2.0数据集是业界规模最大的基于schema的中文信息抽取数据集,其包含超过21万中文句子及48个已定义好的schema,表1 中展示了DuIE2.0数据集中包含的43个简单知识的schema及对应的例子,表2 中展示了DuIE2.0数据集中包含的5个复杂知识的schema及对应的例子。数据集中的句子来自百度百科、百度贴吧和百度信息流文本。数据集划分为17万训练集,2万验证集和2万测试集。其中训练集和验证集用于训练,可供自由下载,测试集分为两个,测试集1供参赛者在平台上自主验证,测试集2在比赛结束前1周发布,并将作为最终的评测排名。`

https://github.com/PaddlePaddle/Research/tree/master/KG/DuIE_Baseline

{"object_type": {"@value": "学校"}, "predicate": "毕业院校", "subject_type": "人物"}
{"object_type": {"@value": "人物"}, "predicate": "嘉宾", "subject_type": "电视综艺"}
{"object_type": {"inWork": "影视作品", "@value": "人物"}, "predicate": "配音", "subject_type": "娱乐人物"}
{"object_type": {"@value": "歌曲"}, "predicate": "主题曲", "subject_type": "影视作品"}
{"object_type": {"@value": "人物"}, "predicate": "代言人", "subject_type": "企业/品牌"}
{"object_type": {"@value": "音乐专辑"}, "predicate": "所属专辑", "subject_type": "歌曲"}
{"object_type": {"@value": "人物"}, "predicate": "父亲", "subject_type": "人物"}
{"object_type": {"@value": "人物"}, "predicate": "作者", "subject_type": "图书作品"}
{"object_type": {"inArea": "地点", "@value": "Date"}, "predicate": "上映时间", "subject_type": "影视作品"}
{"object_type": {"@value": "人物"}, "predicate": "母亲", "subject_type": "人物"}
{"object_type": {"@value": "Text"}, "predicate": "专业代码", "subject_type": "学科专业"}
{"object_type": {"@value": "Number"}, "predicate": "占地面积", "subject_type": "机构"}
{"object_type": {"@value": "Text"}, "predicate": "邮政编码", "subject_type": "行政区"}
{"object_type": {"inArea": "地点", "@value": "Number"}, "predicate": "票房", "subject_type": "影视作品"}
{"object_type": {"@value": "Number"}, "predicate": "注册资本", "subject_type": "企业"}
{"object_type": {"@value": "人物"}, "predicate": "主角", "subject_type": "文学作品"}
{"object_type": {"@value": "人物"}, "predicate": "妻子", "subject_type": "人物"}
{"object_type": {"@value": "人物"}, "predicate": "编剧", "subject_type": "影视作品"}
{"object_type": {"@value": "气候"}, "predicate": "气候", "subject_type": "行政区"}
{"object_type": {"@value": "人物"}, "predicate": "歌手", "subject_type": "歌曲"}
{"object_type": {"inWork": "作品", "onDate": "Date", "@value": "奖项", "period": "Number"}, "predicate": "获奖", "subject_type": "娱乐人物"}
{"object_type": {"@value": "人物"}, "predicate": "校长", "subject_type": "学校"}
{"object_type": {"@value": "人物"}, "predicate": "创始人", "subject_type": "企业"}
{"object_type": {"@value": "城市"}, "predicate": "首都", "subject_type": "国家"}
{"object_type": {"@value": "人物"}, "predicate": "丈夫", "subject_type": "人物"}
{"object_type": {"@value": "Text"}, "predicate": "朝代", "subject_type": "历史人物"}
{"object_type": {"inWork": "影视作品", "@value": "人物"}, "predicate": "饰演", "subject_type": "娱乐人物"}
{"object_type": {"@value": "Number"}, "predicate": "面积", "subject_type": "行政区"}
{"object_type": {"@value": "地点"}, "predicate": "总部地点", "subject_type": "企业"}
{"object_type": {"@value": "地点"}, "predicate": "祖籍", "subject_type": "人物"}
{"object_type": {"@value": "Number"}, "predicate": "人口数量", "subject_type": "行政区"}
{"object_type": {"@value": "人物"}, "predicate": "制片人", "subject_type": "影视作品"}
{"object_type": {"@value": "Number"}, "predicate": "修业年限", "subject_type": "学科专业"}
{"object_type": {"@value": "城市"}, "predicate": "所在城市", "subject_type": "景点"}
{"object_type": {"@value": "人物"}, "predicate": "董事长", "subject_type": "企业"}
{"object_type": {"@value": "人物"}, "predicate": "作词", "subject_type": "歌曲"}
{"object_type": {"@value": "作品"}, "predicate": "改编自", "subject_type": "影视作品"}
{"object_type": {"@value": "企业"}, "predicate": "出品公司", "subject_type": "影视作品"}
{"object_type": {"@value": "人物"}, "predicate": "导演", "subject_type": "影视作品"}
{"object_type": {"@value": "人物"}, "predicate": "作曲", "subject_type": "歌曲"}
{"object_type": {"@value": "人物"}, "predicate": "主演", "subject_type": "影视作品"}
{"object_type": {"@value": "人物"}, "predicate": "主持人", "subject_type": "电视综艺"}
{"object_type": {"@value": "Date"}, "predicate": "成立日期", "subject_type": "机构"}
{"object_type": {"@value": "Text"}, "predicate": "简称", "subject_type": "机构"}
{"object_type": {"@value": "Number"}, "predicate": "海拔", "subject_type": "地点"}
{"object_type": {"@value": "Text"}, "predicate": "号", "subject_type": "历史人物"}
{"object_type": {"@value": "国家"}, "predicate": "国籍", "subject_type": "人物"}
{"object_type": {"@value": "语言"}, "predicate": "官方语言", "subject_type": "国家"}
相关推荐
©️2020 CSDN 皮肤主题: 技术黑板 设计师:CSDN官方博客 返回首页