基于BILSTM_CRF的知识图谱实体抽取方法  

KNOWLEDGE GRAPH ENTITY EXTRACTION BASED ON BILSTM_CRF

作　　者：翟社平 段宏宇 李兆兆 Zhai Sheping;Duan Hongyu;Li Zhaozhao(School of Computer Science and Technology,Xi’an University of Posts and Telecommunications,Xi’an 710121,Shaanxi,China;Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi’an 710121,Shaanxi,China)

机构地区：[1]西安邮电大学计算机学院,陕西西安710121 [2]陕西省网络数据分析与智能处理重点实验室,陕西西安710121

出　　处：《计算机应用与软件》2019年第5期269-274,280,共7页Computer Applications and Software

基　　金：工业和信息化部通信软科学项目(2018-R-26);陕西省自然科学基金资助项目(2012JM8044);陕西省社会科学基金资助项目(2016N008);陕西省教育厅科学研究计划资助项目(12JK0733);西安邮电大学研究生创新基金项目(CXL2016-13)

摘　　要：针对传统知识图谱实体抽取方法需要大量人工特征和专家知识的问题,提出一种基于BILSTM_CRF模型的神经网络结构实体抽取方法。它既能使用双向长短时记忆网络BILSTM(Bidirectional Long Short-Term Memory)提取文本信息的特征,又可利用条件随机场CRF(Conditional Random Fields)衡量序列标注的联系。该方法对输入的文本进行建模,把句子中的每个词转换为词向量;利用BILSTM处理分布式向量得到句子特征;使用CRF标注并抽取实体,得到最终结果。实验结果表明,该方法的准确率和召回率更高,F1值提升约8%,具有更强的适用性。In order to solve the problem that traditional knowledge atlas entity extraction method needed a lot of artificial features and expert knowledge,we proposed a neural network entity extraction method based on BILSTM_CRF model.It could use bidirectional long short-term memory(BILSTM)to extract the features of text information,and use conditional random fields(CRF)to measure the association of sequence labeling.In this method,the input text was modeled,and every word in the sentence was transformed into a word embedding.The distributed vectors were processed by BILSTM to get sentence features.And the CRF tagging and entity extraction were used to get the final results.The experimental results show that the proposed method has higher accuracy and recall rate,and the value of F1 is increased by about 8%,which has better applicability.

关 键 词：知识图谱 实体抽取 神经网络 词向量 BILSTM_CRF模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]