Zhang, ChongHao (Tongji University), Yu, Hao (TongJi University), Luo, Xiao (Tongji University), Yin, Wenyu (Tongji University), Huang, Jinyi (Tongji university), Liu, Xuanyu (Tongji University), Liu, Zhe (Tongji University)

CitySense RAG: Personalized Urban Mobility Recommendations Via Streetscape Perception and Multi-Source Semantics

Scheduled for presentation during the Regular Session "S02b-Optimization for Shared, Electric, and Sustainable Mobility Systems" (WE-EA-T2), Wednesday, November 19, 2025, 13:50−14:10, Southport 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Abstract

With the increasing complexity of urban environments and diversification of user needs, traditional single-modality data processing approaches have revealed significant limitations in personalized recommendation tasks. Existing multimodal Retrieval-Augmented Generation (RAG) systems often focus on integrating only a limited number of modalities, failing to achieve unified representations that encompass street-view imagery, Point of Interest (POI) information, and geographic locations. This study proposes CitySense RAG, an innovative multimodal spatial RAG framework that constructs unified multimodal embeddings through the deep integration of street-view imagery, POI data, and geographic spatial information. this study design an embedding mechanism that fuses spatial proximity with multimodal features, enabling efficient matching between personalized queries and urban spatial data. Experimental results demonstrate that our proposed method outperforms existing techniques in both matching accuracy and semantic expressiveness, achieving an overall performance of 87.3%. Ablation studies further confirm that the integration of the three core components is indispensable for enhanced spatial understanding and recommendation effectiveness.