融合多元用户特征和内容特征的微博谣言实时检测模型

Weibo Rumors Real-time Detection Model Based on Fusion of Multi User Features and Content Features

针对目前基于单文本语义特征深度学习的微博谣言实时检测模型泛化能力不足的问题,提出一种融合多元用户特征和内容特征的实时检测模型.首先,在传统用户基本特征和内容统计特征的基础上,利用用户的历史行为数据,挖掘用户理性值和用户专业度两个深层次特征;然后,基于词向量和带有注意力机制的双向GRU神经网络构建文本语义特征学习模型;最后,采用分层特征级联和全连接的方式进行特征融合,把融合特征输入分类模型进行训练.实验结果表明,该模型的检测准确率达到了91.74%,相比其他只关注文本语义特征的深度学习实时检测模型具有更好的识别效果,相比于其他改进型的实时检测模型F1-Measure值也提高了2.19%.

In order to solve the problem of low generalization ability of Weibo rumor real-time detection model based on deep learning of text semantic features,a real-time detection model fusion of multi user features and content features is proposed.First of all,in addition to the traditional user basic features and content statistical features,two implicit features of user rationality and user professionalism are mined based on user′s historical behavior data;Then,a text semantic feature learning model is constructed based on word vector and bidirectional GRU neural network with attention mechanism;Finally,the hierarchical cascade and full connection are used for feature fusion,and the fused features are input into the classification model for training.The experimental results show that the accuracy of the model is 91.74%,which is better than other deep learning real-time detection models based on text semantic features.Compared with other improved real-time detection models,F1-Measure value also increased by 2.19%.