MobIE

MobIE 是一个德语数据集，它使用 20 种粗粒度和细粒度实体类型以及地理可链接实体的实体链接信息进行人工注释。该数据集由 3,232 个社交媒体文本和带有 91K 标记的交通报告组成，并包含 20.5K 注释实体，其中 13.1K 链接到知识库。数据集的一个子集使用七种与移动相关的 n 元关系类型进行人工注释，而其余文档使用 Snorkel 框架实现的弱监督标签方法进行注释。该数据集可用于 NER（命名实体识别）、EL（实体链接）和 RE（关系提取），因此可用于这些基本信息提取任务的联合和多任务学习。MobIE is a German-language dataset manually annotated with 20 coarse-grained and fine-grained entity types, alongside entity linking information for geographically linkable entities. It consists of 3,232 social media texts and traffic reports with a total of 91K tokens, and contains 20.5K annotated entities, 13.1K of which are linked to knowledge bases. A subset of this dataset was manually annotated using seven mobility-related n-gram relational types, while the remaining documents were annotated through weak supervision labeling methods implemented with the Snorkel framework. This dataset is applicable to NER (Named Entity Recognition), EL (Entity Linking) and RE (Relation Extraction), enabling joint and multi-task learning for these fundamental information extraction tasks.

搜集汇总

背景与挑战

背景概述

MobIE是一个德语数据集，包含3,232个社交媒体文本和交通报告，带有91K标记和20.5K注释实体，其中13.1K链接到知识库。该数据集支持命名实体识别、实体链接和关系提取任务，适用于基本信息提取的联合和多任务学习场景。