JacRED

JacRED是东京工业大学构建的日语文档级关系抽取数据集，旨在促进非英语语言的关系抽取研究。该数据集包含42,241条关系三元组，来源于日文维基百科，通过机器翻译和人工校正相结合的方式构建。JacRED不仅用于评估现有关系抽取模型的性能，还用于探索跨语言关系抽取的挑战，特别是在日语和英语之间的差异性。

JacRED is a Japanese document-level relation extraction dataset developed by Tokyo Institute of Technology, aiming to promote relation extraction research for non-English languages. This dataset contains 42,241 relation triples sourced from Japanese Wikipedia, and is constructed via a combined approach of machine translation and manual correction. JacRED can be used not only to evaluate the performance of existing relation extraction models, but also to explore the challenges of cross-lingual relation extraction, particularly the disparities between Japanese and English.

搜集汇总

背景与挑战

背景概述

JacRED是东京工业大学构建的日语文档级关系抽取数据集，包含42,241条来源于日文维基百科的关系三元组，通过机器翻译和人工校正构建，旨在评估模型性能并探索日语与英语之间的跨语言关系抽取挑战。