基于关系图卷积神经网络的多标签事件预测

Multi-label Event Prediction Based on Relational Graph Convolutional Neural Network

事件预测需要综合考虑的要素众多,现有预测模型多数存在数据稀疏、事件的组合特征及时序特征考虑不足、预测类型单一等问题。为此,提出了基于关系图卷积神经网络的多标签事件预测方法,通过节点特征聚合技术实现数据的稠密化表示。模型利用卷积神经网络的卷积和池化运算,提取预测数据的组合时间段特征信息,并结合长短期记忆网络的时序特征提取能力,进一步提取预测数据的时序规律特征;最后,模型通过全连接的多标签分类器,输出多种类型事件发生的概率值。实验结果表明,所提模型可以支持进行多日期、多类型事件预测,在特定数据集上最高F 1值可以达到0.85。

Event prediction requires comprehensive consideration of many factors,and lots of the existing prediction models have some problems,such as sparse data,insufficient consideration of combined features and sequential features of the events,and single prediction type.To solve these problems,a multi-label event prediction method based on relational graph convolutional neural network is proposed by node feature aggregation technology.The proposed model extracts the combination features by using convolution and pooling algorithm,and extracts the sequence features by using long short term memory networks.Finally,the model outputs the probability for all of the predicted events through a fully connected multi-label classifier.The experimental results show that,the proposed model supports predicting multi-date and multi-type events.The maximum F1 value can achieve 0.85 on specific data set.