Zihao Yu
ABSTRACT
With the widespread application of vehicle GPS equipment, more and more trajectory data are becoming available. Traditional research on trajectory-based route planning uses trajectory data to build a weighted graph, and then obtains routes using classic graph theory-based search algorithms, which has difficulty in estimating multiple preferences. The Recurrent Graph Network (RGN) [1] is a graph neural network, which solves the shortest path problem as a classification problem, i.e., given arbitrary (source, destination) pair, classifying whether the nodes and edges in the graph should be labeled as "on the shortest path connecting the source and destination". Since the RGN model has non-linear feature representation ability and multiple feature fusion ability, this paper considers modifying it for the trajectory-based route planning. However, the actual trajectory data is highly sparse and non-uniform distributed, so that learning trajectory routing in large road networks is a few-shot learning task. To address the aforementioned challenges, this paper proposes the Long Short-Term Memory Graph Network (LSTM-GN) model. The LSTM-GN utilizes a deep LSTM-style message passing process to update various features including comprehensive node-level features, edge-level features and graph-level features, achieving trajectory routing patterns representation based on fewer samples. Additionally, the LSTM-GN needs to get over the difficulty of severe class imbalance and high computational complexity. This paper addresses the above challenges from two perspectives. Firstly, this paper proposes the subgraph filtering algorithms to narrow learning space. Secondly, this paper designs a dynamic weighted loss function to strengthen the positive class. This paper conducts experiments on the real-world trajectory dataset, and the main evaluation metric F1-score of LSTM-GN exceeds the RGN model by 19.18%.
- Battaglia P W, Hamrick J B, Bapst V, et al. Relational inductive biases, deep learning, and graph networks [J]. arXiv preprint arXiv:1806.01261, 2018.Google Scholar
- Guo C, Yang B, Hu J, et al. Learning to route with sparse trajectory sets[C]//2018 IEEE 34th International Conference on Data Engineering (ICDE). IEEE, 2018: 1073--1084.Google Scholar
- Dijkstra E W. A note on two problems in connexion with graphs [J]. Numerische mathematik, 1959, 1(1): 269--271.Google Scholar
- Chen Z, Shen H T, Zhou X. Discovering popular routes from trajectories[C]//2011 IEEE 27th International Conference on Data Engineering. IEEE, 2011: 900--911.Google Scholar
- Wu N, Wang J, Zhao W X, et al. Learning to Effectively Estimate the Travel Time for Fastest Route Recommendation[C]//Proceedings of the 28th ACM International Conference on Information and Knowledge Management. 2019: 1923--1932.Google Scholar
- Feng L, Apers P M G, Jonker W. Towards context-aware data management for ambient intelligence[C]//International conference on database and expert systems applications. Springer, Berlin, Heidelberg, 2004: 422--431.Google Scholar
- Gilmer J, Schoenholz S S, Riley P F, et al. Neural message passing for quantum chemistry[C]//Proceedings of the 34th International Conference on Machine Learning-Volume 70. JMLR. org, 2017: 1263--1272.Google Scholar
- Hochreiter S, Schmidhuber J. Long short-term memory [J]. Neural computation, 1997, 9(8): 1735--1780.Google Scholar
- Guo C, Yang B, Andersen O, et al. Ecomark 2.0: empowering eco-routing with vehicular environmental models and actual vehicle fuel consumption data [J]. GeoInformatica, 2015, 19(3): 567--599.Google ScholarDigital Library
- Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2980--2988.Google Scholar
- Meert W, Verbeke M. HMM with non-emitting states for Map Matching[C]//European Conference on Data Analysis (ECDA), Date: 2018/07/04-2018/07/06, Location: Paderborn, Germany. 2018.Google Scholar
- Smith L N. Cyclical learning rates for training neural networks[C]//2017 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, 2017: 464--47.Google Scholar
Index Terms
- Learning Trajectory Routing with Graph Neural Networks
Recommendations
Dual-view hypergraph neural networks for attributed graph learning
AbstractGraph embedding analyzes network data by learning the vector representation of each vertex in the network, and has attracted widespread attention in recent years. In many real-world networks, considering the topology and attributes of ...
Learning Graph Neural Networks on Feature-Missing Graphs
Knowledge Science, Engineering and ManagementAbstractGraph neural networks have demonstrated state-of-the-art performance in many graph analysis tasks. However, relying on both node features and topology completeness can be challenging, especially as node features may be completely missing. Existing ...
Saliency-aware regularized graph neural network
AbstractThe crux of graph classification lies in the effective representation learning for the entire graph. Typical graph neural networks focus on modeling the local dependencies when aggregating features of neighboring nodes, and obtain the ...
Comments