基于卷积神经网络的谣言检测

发布时间：2018-03-13 02:38

本文选题：微博　切入点：谣言检测　出处：《计算机应用》2017年11期 　论文类型：期刊论文

【摘要】：人工检测谣言通常需要耗费大量的人力物力,并且会有很长的检测延迟。目前现存的谣言检测模型一般根据谣言的内容、用户属性、传播方式人工地构造特征,而人工构建特征存在考虑片面、浪费人力等现象。为了解决这个问题,提出了基于卷积神经网络(CNN)的谣言检测模型。将微博中的谣言事件向量化,通过卷积神经网络隐含层的学习训练来挖掘表示文本深层的特征,避免了特征构建的问题,并能发现那些不容易被人发现的特征,从而产生更好的效果。实验结果表明,所提方法能够准确识别谣言事件,在准确率、精确率与F1值指标上优于支持向量机(SVM)与循环神经网络(RNN)等对比算法。
[Abstract]:Artificial detection rumors usually consumes a lot of manpower and resources, and there will be a long delay detection. At present the existing rumor detection model according to the general contents of rumors, user properties, mode of transmission of artificial structures, and constructed considering the existing characteristics of one-sided, waste of human resources and so on. In order to solve this problem, put forward convolutional neural network (CNN) based on rumor detection model. Micro-blog rumors in event vectorization, through convolution neural network hidden layer training to represent text mining deep features, to avoid the characteristics of construction problems, and found that those who are not easily found features, resulting in a better effect. The experimental results show that the proposed method can accurately identify the rumor event in accuracy, the accurate rate of indicators is better than the support vector machine and F1 value (SVM) and recurrent neural network (RNN) comparison algorithm.

【作者单位】：同济大学计算机科学与技术系;嵌入式系统与服务计算教育部重点实验室(同济大学);
【基金】：国家自然科学基金资助项目(61573259,61673301,61573255,61673299) 上海市中医药三年行动计划重点项目(ZY3-CCCX-3-6002) 上海自然科学基金资助项目(15ZR1443800)~~
【分类号】：TP183;TP393.092

【相似文献】