一种基于CNN-RNN的社交媒体中突发事件感知方法

Emergency Detection Method in Social Media Based on CNN-RNN

针对传统社交媒体信息感知中过度依赖语义及社会网络信息的问题,提出一种基于CNN与RNN融合的社交媒体中突发事件的感知方法。该方法在定义社交媒体中时空语义问题的基础上,构建融合CNN-RNN的突发事件感知模型,通过多层CNN网络对社交消息按照主题进行聚类,然后将同一主题的消息按照时间序列输入RNN循环网络单元,当事件消息传播到一定范围或一定时长时,RNN输出相应突发事件主题,并通过地理位置匹配找到现场用户发出的源消息,对这些源消息进行事件画像以实现对突发事件的感知。实验结果表明,与传统基于语义和社会网络信息的方法相比,本文方法不需要构建复杂的社会网络模型,不依赖于语义信息检索,F 1值达到93.4%,能够快速实现对突发事件的精准感知。

In response to the problem of excessive reliance on semantics and social network information in traditional social media information detection,a method for detecting emergencies in social media based on the fusion of convolutional neural network(CNN)and recurrent neural network(RNN)was proposed.The emergent event perception model fused with CNN-RNN was constructed on the basis of defining temporal and spatial semantic issues in social media.Social messages were clustered by topic through multi-layer CNN network.Then the messages of the same topic were input into the RNN recurrent network unit in time series.When the event message spread to a certain range or duration,the RNN output the corresponding emergency topic and found the source message sent by the on-site users through the geographical location matching.The scene of emergencies was quickly perceived by portraying these source messages.The results show that compared with traditional detection methods based on semantics and social network information,the proposed method does not need to construct a complex social network model nor does it rely on semantic information retrieval.With the F 1 value reaching 0.934,emergency events can be quickly detected out.