基于优化长短期记忆网络的数字资源关系抽取研究  

Research on relationship extraction of digital resources based on optimized short-term memory network

作　　者：费晶[1] FEI Jing(Yulin university,Yulin Shaanxi,719000,China)

机构地区：[1]榆林学院,陕西榆林719000

出　　处：《自动化与仪器仪表》2023年第3期63-66,71,共5页Automation & Instrumentation

基　　金：榆林市2022年科技计划《图书馆古籍文献数字人文开发的应用研究》(CXY-2022-97)。

摘　　要：数字资源存在内容繁杂、结构混乱等问题影响了信息的快速抽取。为了提高其有效利用率,此次实验改进了数据预处理方法来降低误差累积,然后选择长短时记忆算法来提取特征,引入注意力机制来优化并建立关系抽取模型。结果显示在ACE-05和Sem Eval-2010Task-8数据集中,输入WF特征的改进算法F1值分别为55.3%和74.7%,输入特征为WF+PF+EI+DEP的F1值为64.4%和84.2%。在Sem Eval-2010Task-8数据集中,改进算法对CC关系的特征提取精确率最高为91.7%,对ED关系的特征提取F1值最高为90.5%,ED关系的特征提取召回率最高为88.8%。可以看出改进后的资源抽取算法具有较好的性能优势,可以应用于数据资源的关系抽取中。The problems of digital resources,such as complicated content and chaotic structure,affect the rapid extraction of information.In order to improve its effective utilization,this experiment improved the data preprocessing method to reduce the error accumulation,and then selected long short-term memory algorithm is used to extract features,and introduced the attention mechanism to optimize and establish the relationship extraction model.The results show that in ACE-05 and Sem Eval-2010Task-8 datasets,the improved algorithm F1 values for WF feature input are 55.3%and 74.7%,respectively,and the F1 values for WF+PF+EI+DEP feature input are 64.4%and 84.2%.In the Sem Eval-2010Task-8 dataset,the improved algorithm has the highest feature extraction accuracy of 91.7%for CC relations,90.5%for ED relations,and 88.8%for ED relations.It can be seen that the improved resource extraction algorithm has better performance advantages and can be applied to relational extraction of data resources.

关 键 词：长短时记忆 数字资源 神经网络 注意力机制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]