摘要
现有研究多采用抽取语法特征构建特征向量的方式完成企业关系抽取任务。但企业文本语法特征复杂,长程依赖明显。根据企业文本的特点,提出基于Bi-GRU网络抽取文本特征、分段词汇层注意力机制及句子层注意力机制从多个维度抽取企业文本的整体特征。实验表明,在公开的上市公司关系数据中,准确率、召回率和F值相较于Bi-LSTM方法有明显的提高。
According to the characteristics of the corporate text,this paper extracts text features based on Bi-GRU network,the segmentation lexical layer attention mechanism and the sentence layer attention mechanism extract the characteristics of corporate text. Experiments show that in the company relationship data,the accuracy rate,recall rate and F value are significantly improved compared to the Bi-LSTM method.
出处
《工业控制计算机》
2019年第7期113-115,共3页
Industrial Control Computer
关键词
企业关系抽取
Bi-GRU
注意力机制
enterprise relationship extraction
Bi-GRU
attention mechanism
