Shiming Zhang a,b,c,d, Zhipeng Luo a, Li Yang a, Fei Teng a,b,c,d,Tianrui Li a,b,c,d a- School of Computing and Artificial Intelligence, Southwest Jiaotong Universityb- EngineeringResearchCenterofSustainableUrbanIntelligentTransportation,Ministry of Education c- National Engineering Laboratory of Integrated Transportation Big Data Application Technology, Southwest Jiaotong University d- Manufacturing Industry Chains Collaboration and Information Support Technology Key Laboratory of Sichuan Province,SouthwestJiaotong University, Chengdu 611756,Sichuan,P.R.China AbstractNowadays,with advanced information technologies deployed city wide, large data volumes and powerful computational resources are intelligentizing modern city development. As an important part of intelligent transportation, route recommendation and its applications are widely used, directly influencing citizens’ travel habits. Developing smartandefficient travel routes based on big data (possiblymulti-modal) has become a central challenge in route recommendation research. Our survey offers a comprehensive review of route recommendation work based on urban computing. It is organized by the following three parts:1) Methodology wise. We categorize a large volume of traditional machine learning and modern deep learning methods. Also, we discuss their historical relations and reveal the edge-cutting progress. 2) Application-wise. We present numerous novel applications related to route commendation within urban computing scenarios. 3) We discuss current problems and challenges and envision several promising research directions.We believe that this survey can help relevant researchers quickly familiarize themselves with the current state of route recommendation research and then direct them to future research trends.
Gao et al. used graph a spatio-temporal network to fuse POI and track location information, and employed transfer learning techniques to address the data sparsity problem in travel route recommendation. Wang et al. [75, 81] used a GNN to estimate the road distance, travel time, and heuristic function h(x) in A* based on the extracted features of the GNN. Wen et al. employed a GCN to encode the checkpoints and trajectories along goods delivery paths and extracted the spatio-temporal dependencies for providing more profitable routes. Wu et al. addressed the traveling salesman problem by using graph attention networks for path node embedding and applied their model for delivery route recommendations. Wu et al. utilized a graph attention network to extract features from road vehicle trajectory and speed information and improved upon the A* constraint method. Wu et al. constructed a three-leve
id: 7f8d27b9a45a927de3c86895533ddefd - page: 12
developed NeuroMLR, which tions. learns a generative model from historical trajectories by conditioning on three explanatory factors: the current location, the destination, and real-time traffic conditions. The conditional distributions are learned through a combination of Lipschitz embedding with a GCN using historical trajectory data. Yang et al. applied graph attention networks and pre-trained models to embed graph-structured data and combined A* to compute approximate graph edit distance in a more effective and intelligent way.
id: 8f69e6a9aa8083f0033692ad328a1de7 - page: 12
As we have seen, GNNs can address some challenges of data representation and information extraction problems in graph-structured data. Compared with sequence-based models that only extract temporal correlations, GNNs can capture both temporal and spatial dependencies in many urban computing tasks, where the models need to process various types of information such as user activities, trajectories, road information, traffic flows, and other physical factors.
id: 35448bffe2d1e74f6b3a7f020dcd70a0 - page: 12
4.5. Multi-modal Approaches Most of the existing route recommendation methods process two types of data trajectories and road network information. However, with the development of urban information systems, many more types of data are recorded, and thus single-modal methods may not be sufficient to meet the travel needs of users. Meanwhile, multi-modal learning paradigms have been shown very 12 effective in other research areas such as computer vision and natural language processing [124, 125]. Therefore, multi-modal route recommendation methods are a promising research area. New modalities in urban computing include but are not limited to weather information, different types of transportation networks, scenic images, texts, and audio. In this sub-section, we review the literature from two perspectives: one mainly fuses images, texts, audio, and weather data (Figure 6), and the other fuses different types of transportation data (Figure 7).
id: 8fa18d3c2569c4b24a1deb6ecfdac00c - page: 12