融合时间序列与卷积神经网络的网络谣言检测  

Network Rumor Detection Combining Time Series and Convolutional Neural Network

作　　者：汪建梅 彭云[1] 余晨钰 WANG Jian-mei;PENG Yun;YU Chen-yu(School of Computer and Information Engineering,Jiangxi Normal University,Nanchang 330022,China)

机构地区：[1]江西师范大学计算机信息工程学院,南昌330022

出　　处：《小型微型计算机系统》2022年第5期1020-1026,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61966017,61662032)资助;江西省高校人文社科项目(JC19121,JC19117)资助。

摘　　要：针对卷积神经网络检测网络谣言没有考虑到谣言各生命周期之间的深层特征以及训练参数过于庞大问题,提出了一种融合时间序列和卷积神经网络的谣言检测算法(CNN-TS).首先将微博中的谣言事件向量化,其次将这些向量根据时间顺序分组为潜伏期、滋生期、蔓延期、消退期,然后通过卷积神经网络隐含层的学习训练来挖掘表示各阶段文本深层的特征,最后通过修改分类函数进一步加强学习效果.实验结果表明,该检测算法在准确率、精确率及F1值上均优于支持向量机与卷积神经网络,能够准确识别谣言事件.Aiming at the problem that the convolutional neural network detects network rumors without considering the deep features between the life cycles of the rumors and the excessive training parameters,a rumor detection algorithm combining time series and convolutional neural networks is proposed.First,the rumors in Weibo are vectorized,and then these vectors are grouped into incubation period,breeding period,spreading period,and fading period according to the time sequence,and then through the learning and training of the hidden layer of the convolutional neural network to mine the deep text of each stage.Finally,the learning effect is further enhanced by modifying the classification function.Experimental results show that the proposed method can accurately identify rumor events and is superior to comparison algorithms such as support vector machines and convolutional neural networks in terms of accuracy,precision and F1index.

关 键 词：卷积神经网络 谣言检测 时间序列 分类函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]