本项目为论文 Knowledge-guided Open Relation Extraction 的源代码,目前该论文还在投稿过程中
This repo contains the source code of the paper "Knowledge-guided Open Relation Extraction". The paper is still in the process of submission
模型的结构如下图所示:
The model's architecture is shown in the figure below:
它使用了Attention机制实现了实体和背景知识与原文信息的融合
It combines the entity and background knowledge with the origin sentence
将COER数据集链接到wikidata获得COER-linked
We link COER to wikidata and obtain the COER-linked
数据集可以通过以下链接下载:
The dataset can be obtained from this link:
链接:https://pan.baidu.com/s/1giJEutSSRPmwfyA1mog1VQ
提取码:8kt3
在COER-linked上,本模型的效果为
The model's performance on COER-linked is
Precision:90.41 Recall:89.99 F1:90.20
执行
python one_attempt.py
以运行5次训练和测试,保存最佳的模型