民航旅客行程状态推断方法研究

现有的民航旅客行程状态推断的相关方法不能对旅客历史行为序列中远距离项的依赖关系建模,且忽略了行程子结构,为此提出一种新的模型来推断旅客行程状态。首先通过图神经网络挖掘出行序列任意机场间的转移模式;其次,构造层次注意力依次在机场级别和行程级别捕获旅客的短期和长期出行偏好;最后融合旅客的短期和长期出行偏好进行分类。实验结果表明,图神经网络的消息传递机制突破了距离的限制,有效捕获了旅客出行序列中任意机场间复杂的关系,模型在多项性能指标上效果很好,构建的图神经网络和注意力机制结合的方法可获得更好的性能;另外,结合实际的应用场景,融入了额外的特征进行信息补充取得了更好的推断效果。

Study on inference for itinerary status of civil aviation passengers

The previous relevant studies fail to model the dependencies between distant items and ignore the substructure of itinerary.To solve this problem,this paper proposed a sequence model to judge the itinerary status of passengers.Firstly,the method used graph neural network to capture the transfer pattern between any airports in the travel sequence.Secondly,it constructed hierarchical attention mechanism to capture the short-term and long-term travel preferences of passengers at the airport level and the itinerary level respectively.Finally,it designed the final classification by combining the long-term and short-term travel preference of passengers.Experimental results show that the message passing mechanism of the graph neural network breaks through the limit of distance,and effectively captures the complex relationship between any items in the passenger travel sequence.The proposed model gets superior performance in real dataset by combining the graph neural network and attention mechanism.In addition,combining with the actual application scenarios,using the additional information gets a better result.