文献类型：期刊文献

中文题名：网络文本蕴含关系识别的异常信息获取仿真

英文题名：Simulation of Abnormal Information Acquisition for Network Text implication Relationship Recognition

作者：赵海燕[1];刘琨[1];王廷梅[1];杜丽娟[1]

第一作者：赵海燕

机构：[1]北京联合大学应用科技学院,北京100012

第一机构：北京联合大学应用科技学院

年份：2020

卷号：37

期号：8

起止页码：256-260

中文期刊名：计算机仿真

外文期刊名：Computer Simulation

收录：CSTPCD;;北大核心:【北大核心2017】；

语种：中文

中文关键词：网络文本;关系识别;异常信息;分段卷积神经网络

外文关键词：Network text;Relation recognition;Abnormal information;Piecewise convolution neural network

摘要：为优化地方志语料库信息完整度及数据提供的可靠性,需及时检测语料库中网络文本间存在的异常信息。为此以网络文物文本蕴含关系作为分析对象,提出基于卷积神经网络算法的文本关系异常信息获取方法。在地方志语料库网络中采集并统计标注信息等知识源,并将其作为文物的文本信息,构建分段卷积神经网络训练获取到的文本信息,进行文本语义空间向量化,并得到文物文本蕴含关系识别结果。基于此选取KNN算法获取网络异常信息,利用截断距离获取的底层网络局部密度与距离获取网络文物文本蕴含关系识别结果中各数据样本点,文物文本信息数据局部密度与间距满足设定阈值时,判定该样本点为异常数据。仿真结果表明,采用上述方法识别文物文本蕴含关系准确率均在97.5%以上,获取网络异常信息准确率均高于98.5%,可应用于网络异常状态监测中。
In order to optimize the information integrity of local chronicle corpus and the reliability of data,it is necessary to detect the abnormal information in time.Therefore,the text implication of network cultural relic was taken as the analysis object and a method for obtaining the abnormal information of text relation based on convolution neural network algorithm.The knowledge source such as annotation was collected and counted in network of local chronicle corpus,and it was taken as the text information of cultural relics.Moreover,the text information trained by piecewise convolution neural network was constructed,and the text semantic space vector was vectorized.After that,the recognition result of text implication of cultural relics was obtained.On this basis,KNN algorithm was selected to obtain the abnormal information.Each data sample point could be obtained by the local density and distance of underlying network obtained by truncation distance.When the local density and spacing of text data met the setting threshold,the sample point was determined as the abnormal data.Simulation results prove that the accuracy of identifying of text implication by the proposed method is more than 97.5%,and the accuracy of abnormal network information is higher than 98.5%,so this method can be applied to abnormal state monitoring.