论文
Relation Classification via Convolutional Deep Neural Network
- 发布年份:2014
- 会议:COLING
- 作者:Daojian Zeng,Kang Liu,Siwei Lai,Guangyou Zhou and Jun Zhao
- 机构:National Laboratory of Pattern Recognition,Institute of Automation
- 贡献:第一次在 relation classification 领域运用 convolutional DNN
- 数据集: SemEval-2010 Task 8 dataset, http://docs.google.com/View?id=dfvxd49s。该数据集包含 10717 条被标注的数据(包括 8000 条训练用例和 2717 条测试用例。
模型结构
整个模型得架构如下:
- Input:带有两个被标记单词的一句话。
- Word Representation:将每个单词转为词嵌入。
- Feature Extraction:分别提取单词级别的特征和句子级别的特征,然后将两个特征向量连接起来。
- Output:简单的线性层+softmax 分类。