Adaptive model selection and ensemble via spatiotemporal graph-guided expert routing
文献类型：期刊论文
| 作者 | Wang, Lizeng3,4; Cheng, Shifen3,4; Lu, Feng1,2,3,4 |
| 刊名 | INFORMATION PROCESSING & MANAGEMENT
 |
| 出版日期 | 2026-11-01 |
| 卷号 | 63期号:7页码:104814 |
| 关键词 | Ensemble learning Spatiotemporal inference Model selection Spatiotemporal correlation Spatiotemporal heterogeneity Mixture of experts |
| ISSN号 | 0306-4573 |
| DOI | 10.1016/j.ipm.2026.104814 |
| 产权排序 | 1 |
| 文献子类 | Article |
| 英文摘要 | Ensemble learning has shown strong potential for spatiotemporal inference by integrating multiple base models to improve accuracy and robustness. However, existing methods often rely on fixed sets of base models and localized ensemble strategies, limiting their adaptability to dynamic data patterns and global spatiotemporal correlations. To address this, we propose an adaptive ensemble framework that dynamically selects and fuses multiple models under varied spatiotemporal conditions. First, a graph-based module is presented to encode the spatial relationships among sensor nodes and the temporal dynamics of their time series into unified context embeddings. Second, an adaptive routing mechanism is designed to compute context-dependent response scores to guide model selection. Finally, a context-specific ensemble strategy aggregates model outputs using weights derived from these scores. Experiments on traffic flow, traffic speed, and air quality datasets show that the proposed framework achieves consistently competitive and often better performance than both mainstream ensemble methods and recent dynamic graph models, reducing inference errors by up to 2.1%-4.7% while using 11%-60% fewer parameters and maintaining comparable inference efficiency. Further interpretability analysis confirms that the framework effectively captures spatiotemporal correlations and performance heterogeneity, thereby enabling adaptive model selection and fusion in response to local contextual variations. |
| URL标识 | 查看原文 |
| WOS研究方向 | Computer Science ; Information Science & Library Science |
| 语种 | 英语 |
| WOS记录号 | WOS:001745583300001 |
| 出版者 | ELSEVIER SCI LTD |
| 源URL | [http://ir.igsnrr.ac.cn/handle/311030/221544]  |
| 专题 | 资源与环境信息系统国家重点实验室_外文论文 |
| 通讯作者 | Cheng, Shifen |
| 作者单位 | 1.Jiangsu Ctr Collaborat Innovat Geog Informat Resou, Nanjing 210023, Peoples R China 2.Fuzhou Univ, Acad Digital China, Fuzhou, Peoples R China; 3.Univ Chinese Acad Sci, Beijing 100049, Peoples R China; 4.Chinese Acad Sci, Inst Geog Sci & Nat Resources Res, State Key Lab Resources & Environm Informat Syst, Beijing 100101, Peoples R China; |
| 推荐引用方式 GB/T 7714 | Wang, Lizeng,Cheng, Shifen,Lu, Feng. Adaptive model selection and ensemble via spatiotemporal graph-guided expert routing[J]. INFORMATION PROCESSING & MANAGEMENT,2026,63(7):104814. |
| APA | Wang, Lizeng,Cheng, Shifen,&Lu, Feng.(2026).Adaptive model selection and ensemble via spatiotemporal graph-guided expert routing.INFORMATION PROCESSING & MANAGEMENT,63(7),104814. |
| MLA | Wang, Lizeng,et al."Adaptive model selection and ensemble via spatiotemporal graph-guided expert routing".INFORMATION PROCESSING & MANAGEMENT 63.7(2026):104814. |
入库方式： OAI收割
来源：地理科学与资源研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。