深度学习在轨迹数据挖掘中的应用研究综述

A Review of the Application of Deep Learning in Trajectory Data Mining

摘要: 在过去十年，深度学习已被证明在很多领域应用非常成功，如视觉图像、自然语言处理、语音识别等，同时也涌现出来了大量深度学习模型，如深度卷积神经网络，深度循环神经网络、深度信念网络、深度对抗生成网络等，其深度越深学习能力越强，但同时也增加了网络训练的复杂度，权衡利弊取其中，所以在不同的应用中有很大的研究空间。另外，在基于时序数据的应用研究中，如股票趋势预测、天气预测、异常气候地质灾害预测等，相比较于传统的方法，各种改进的、融合的深度学习方法显然更胜一筹。而轨迹数据是时序数据的一种，它除了有时间维度的依赖性还有空间维度的相关关系。轨迹数据挖掘与我们的生活息息相关，从城市规划到个性化推荐，从出行安全到优质服务。所以本文通过对轨迹数据进行简单的剖析，对轨迹数据的应用进行一定的总结，并分析了一些用于轨迹数据挖掘的深度学习模型的优缺点，提出了一些小技巧，以此来对后续轨迹数据挖掘提供一些思路。

Abstract: In the last decade, deep learning has been proved to be successful in many fields, like visual images, natural language processing and speech recognition. At the same time, a plenty of deep learning models showed their powerful learning abilities, such as Deep Convolutional Neural Network, Deep Recurrent Neural Network, Deep Belief Network and Deep Generative Adversarial Network. The deeper the models designed, the better the capability of learning would be, but more complex the training process would become. Therefore, to find a balanced pointinan application would be significant. And there will be much space for research. In addition, it also achieved state-of-the-art performance in time series data applications, like stock trend prediction, weather forecast, abnormal climate and extreme geological disaster prediction, etc. Specially, trajectory data not only has the time dimension relationship but is also related in spatial dimension. And it is closely related to our daily life, from urban planning to individual recommendation, from safe travel to qualified services. In this paper, we briefly introduced the trajectory data, summarized some applications in trajectory data mining and analyzed the advantages and disadvantages on the frequently used deep learning models in trajectory data mining, and put forward some small tricks to help with some new thoughts to the later research in this field.