File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Conference Paper: Spatial-Temporal Convolutional Graph Attention Networks for Citywide Traffic Flow Forecasting

TitleSpatial-Temporal Convolutional Graph Attention Networks for Citywide Traffic Flow Forecasting
Authors
Keywordsgraph attention networks
spatial-temporal data mining
traffic flow forecasting
urban computing
Issue Date2020
Citation
International Conference on Information and Knowledge Management, Proceedings, 2020, p. 1853-1862 How to Cite?
AbstractTraffic flow prediction plays an important role in many spatial-temporal data applications, e.g., traffic management and urban planning. Various deep learning techniques are developed to model the traffic dynamic patterns with different neural network architectures, such as attention mechanism, recurrent neural network. However, two important challenges have yet to be well addressed: (i) Most of these methods solely focus on local spatial dependencies and ignore the global inter-region dependencies in terms of traffic distributions; (ii) It is important to capture channel-aware semantics when performing spatial-temporal information aggregation. To address these challenges, we propose a new traffic prediction framework - Spatial-Temporal Convolutional Graph Attention Network (ST-CGA), to enable the traffic prediction with the modeling of region dependencies, from locally to globally in a comprehensive manner. In our ST-CGA framework, we first develop a hierarchical attention networks with a graph-based neural architecture, to capture both the multi-level temporal relations and cross-region traffic dependencies. Furthermore, a region-wise spatial relation encoder is proposed to supercharge ST-CGA mapping spatial and temporal signals into different representation subspaces, with channel-aware recalibration residual network. Extensive experiments on four real-world datasets demonstrate that ST-CGA achieve substantial gains over many state-of-the-art baselines. Source codes are available at: https://github.com/shurexiyue/ST-CGA.
Persistent Identifierhttp://hdl.handle.net/10722/308829
ISI Accession Number ID

 

DC FieldValueLanguage
dc.contributor.authorZhang, Xiyue-
dc.contributor.authorHuang, Chao-
dc.contributor.authorXu, Yong-
dc.contributor.authorXia, Lianghao-
dc.date.accessioned2021-12-08T07:50:13Z-
dc.date.available2021-12-08T07:50:13Z-
dc.date.issued2020-
dc.identifier.citationInternational Conference on Information and Knowledge Management, Proceedings, 2020, p. 1853-1862-
dc.identifier.urihttp://hdl.handle.net/10722/308829-
dc.description.abstractTraffic flow prediction plays an important role in many spatial-temporal data applications, e.g., traffic management and urban planning. Various deep learning techniques are developed to model the traffic dynamic patterns with different neural network architectures, such as attention mechanism, recurrent neural network. However, two important challenges have yet to be well addressed: (i) Most of these methods solely focus on local spatial dependencies and ignore the global inter-region dependencies in terms of traffic distributions; (ii) It is important to capture channel-aware semantics when performing spatial-temporal information aggregation. To address these challenges, we propose a new traffic prediction framework - Spatial-Temporal Convolutional Graph Attention Network (ST-CGA), to enable the traffic prediction with the modeling of region dependencies, from locally to globally in a comprehensive manner. In our ST-CGA framework, we first develop a hierarchical attention networks with a graph-based neural architecture, to capture both the multi-level temporal relations and cross-region traffic dependencies. Furthermore, a region-wise spatial relation encoder is proposed to supercharge ST-CGA mapping spatial and temporal signals into different representation subspaces, with channel-aware recalibration residual network. Extensive experiments on four real-world datasets demonstrate that ST-CGA achieve substantial gains over many state-of-the-art baselines. Source codes are available at: https://github.com/shurexiyue/ST-CGA.-
dc.languageeng-
dc.relation.ispartofInternational Conference on Information and Knowledge Management, Proceedings-
dc.subjectgraph attention networks-
dc.subjectspatial-temporal data mining-
dc.subjecttraffic flow forecasting-
dc.subjecturban computing-
dc.titleSpatial-Temporal Convolutional Graph Attention Networks for Citywide Traffic Flow Forecasting-
dc.typeConference_Paper-
dc.description.naturelink_to_subscribed_fulltext-
dc.identifier.doi10.1145/3340531.3411941-
dc.identifier.scopuseid_2-s2.0-85095866227-
dc.identifier.spage1853-
dc.identifier.epage1862-
dc.identifier.isiWOS:000749561301086-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats