计算机科学 ›› 2023, Vol. 50 ›› Issue (11): 114-121.doi: 10.11896/jsjkx.221000058
陈嘉俊, 陈伟, 赵雷
CHEN Jiajun, CHEN Wei, ZHAO Lei
摘要: 现有路网场景下的轨迹表示学习(Trajectory Representation Learning,TRL) 方法可分为两类,即基于循环神经网络(RNN)和长短期记忆(LSTM)的序列化模型以及基于自注意力机制的学习模型。尽管已有研究做出了重大贡献,但它们仍然存在以下问题:(1)现有的路网表示学习方法忽略了相邻路段之间的转移概率,不能充分捕获路网的拓扑结构信息;(2)基于自注意力机制的学习模型在短轨迹和中长轨迹上的表现优于序列化模型,但在长轨迹的表示学习上性能较差,未能很好刻画轨迹的长期语义特征。基于此,文中提出了一个新的轨迹表示学习模型TRMS。该模型采用概率感知游走来优化传统DeepWalk算法,以深入挖掘路网的拓扑结构,然后将自注意力机制和Masked Seq2Seq学习框架相结合来捕获轨迹的长期语义特征。最后,基于真实轨迹数据进行实验,结果表明,TRMS在短、中、长轨迹的嵌入表示上,性能都优于最好的基线方法。
中图分类号:
[1]WANG S,BAO Z,XIE Z,et al.Torch:A search engine for tra-jectory data[C]//Proceedings of SIGIR.ACM,2018:535-544. [2]YOU D.Trajectory Pattern Construction and Next LocationPrediction of Individual Human Mobility with Deep Learning Models [J].Computing in Science and Engineering,2020,14(2):52-65. [3]ZHAO J,XU J.On prediction of user destination by sub-trajectory understanding:A deep learning based approach[C]//Proceedings of CIKM.ACM,2018:1413-1422. [4]LI X.Deep representation learning for trajectory similarity computation[C]//Proceedings of ICDE.IEEE,2018:617-628. [5]ENCISO-RODAS L.Trajectory anomaly detection based on simi-larity analysis[C]//Proceedings of CLEI.IEEE,2021:1-10. [6]WANG S,BAO Z,CONG G,et al.A survey on trajectory data management,analytics,and learning [J].ACM Computing Surveys,2021,54(2):1-36. [7]ZHENG Y,LI Q,CHEN Y,et al.Understanding mobility based on GPS data[C]//Proceedings of UbiComp.ACM,2008:312-321. [8]ZHENG Y,LIU L,WANG L,et al.Learning transportationmode from raw GPS data for geographic applications on the web[C]//Proceedings of WWW.ACM,2008:247-256. [9]WU H,CHEN Z,SUN W,et al.Modeling trajectories with recurrent neural networks[C]//Proceedings of IJCAI.IEEE,2017:3083-3090. [10]FU T,LEE W.Trembr:Exploring road networks for trajectory representation learning [J].ACM Transactions on Intelligent Systems and Technology,2020,11(1):1-25. [11]CHEN Y,CONG G,LIU Y,et al.Robust road network representation learning:When traffic patterns meet traveling semantics[C]//Proceedings of CIKM.ACM,2021:211-220. [12]WU N,ZHAO W,WANG J,et al.Learning effective road network representation with hierarchical graph neural networks[C]//Proceedings of KDD.ACM,2020:6-14. [13]JEPSEN T,JENSEN C.Graph convolutional networks for road networks[C]//Proceedings of SIGSPATIAL.ACM,2019:460-463. [14]LEE W,DU Y.Learning embeddings of intersections on road networks[C]//Proceedings of SIGSPATIAL.ACM,2019:309-318. [15]PEROZZI B,SKIENA S.Deepwalk:online learning of socialrepresentations[C]//Proceedings of KDD.ACM,2014:701-710. [16]SONG K,TAN X,QIN T,et al.MASS:masked sequence to sequence pre-training for language generation[C]//Proceedings of ICML.JMLR,2019:5926-5936. [17]GROVER A.node2vec:Scalable feature learning for networks[C]//Proceedings of SIGKDD.ACM,2016:855-864. [18]MIKOLOV T.Distributed representations of sentences and do-cuments[C]//Proceedings of ICML.JMLR,2014:1188-1196. [19]CHANG M,LEE K.BERT:pre-training of deep bidirectionaltransformers for language understanding[C]//Proceedings of NAACL-HLT.ACL,2019:4171-4186. [20]VASWANI A,SHAZEER N,JONES L.Attention is all youneed[C]//Proceedings of NIPS.ACM,2017:5998-6008. [21]RANU S,TELANG A.Indexing and matching trajectories under inconsistent sampling rates[C]//Proceedings of ICDE.IEEE,2015:999-1010. [22]ZHANG C.Map-matching for low-sampling-rate GPS trajectories[C]//Proceedings of SIGSPATIAL.ACM,2009:352-361. [23]YANG C.Enhanced Map-Matching Algorithm with a HiddenMarkov Model for Mobile Phone Positioning[J].International Journal of Geographical Information Science,2017,6(11):327-343. [24]VULIC I.Hello,it's GPT-2-how can I help you?towards theuse of pretrained language models for task-oriented dialogue systems[C]//Proceedings of IJCNLP.ACL,2019:15-22. |
|