Junxiang Ji
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Junxiang Ji has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Longbiao Chen
Xiamen University
6 shared publications - Cheng Wang
Fujian Key Laboratory of Sensing and Computing for Smart Cities, Xiamen University, Xiamen, Fujian, China
4 shared publications - Ding He
Xiamen University
3 shared publications - Guofeng Luo
Xiamen University
3 shared publications - Yongyi Wu
Fujian Key Laboratory of Sensing and Computing for Smart Cities, Xiamen University, Xiamen, China and School of Informatics, Xiamen University, Xiamen, China
3 shared publications - Yufei Wang
Fujian Key Laboratory of Sensing and Computing for Smart Cities, Xiamen University, Xiamen, China and School of Informatics, Xiamen University, Xiamen, China
3 shared publications
Latest Publications
- Open Access
- 2 December 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (4) , 1-32
TourismMinds: A Geo-augmented LLM Framework for Semantic-aware Trajectory Analytics and Generation
With the rise of smart cities and digital tourism, understanding tourist movement is essential for enhancing personalized experiences and informing business decisions. Traditional methods struggle to integrate multi-source data and interpret complex behaviors, while large language models (LLMs) offer untapped potential. This study proposes an LLM-driven “Perception-Modeling-Generation” framework, validated by a Kulangsu smart tourism case. First, a five-stage spatiotemporal-semantic alignment pipeline integrates GPS, social media, POI, and weather data. Second, to address LLM limitations, we introduce a knowledge-enhanced framework combining geospatially-aware LoRA fine-tuning and a hierarchical Retrieval-Augmented Generation (RAG) mechanism. This approach reduces hallucinations by 15% and boosts geographic reasoning accuracy, achieving state-of-the-art intent recognition (15% gain) and landmark identification (72% accuracy). Third, we design a reprogramming strategy with geo-semantic constraints for generating spatially plausible, personalized trajectories in data-scarce settings, showing 89% similarity to real-world data. Supported by the Kulangsu Smart Tourism System, our method provides a scalable paradigm for spatiotemporal modeling and synthetic trajectory generation, benefiting personalized services and decision-making.
- 2 December 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (4) , 1-32
Open AccessTourismMinds: A Geo-augmented LLM Framework for Semantic-aware Trajectory Analytics and Generation
With the rise of smart cities and digital tourism, understanding tourist movement is essential for enhancing personalized experiences and informing business decisions. Traditional methods struggle to integrate multi-source data and interpret complex behaviors, while large language models (LLMs) offer untapped potential. This study proposes an LLM-driven “Perception-Modeling-Generation” framework, validated by a Kulangsu smart tourism case. First, a five-stage spatiotemporal-semantic alignment pipeline integrates GPS, social media, POI, and weather data. Second, to address LLM limitations, we introduce a knowledge-enhanced framework combining geospatially-aware LoRA fine-tuning and a hierarchical Retrieval-Augmented Generation (RAG) mechanism. This approach reduces hallucinations by 15% and boosts geographic reasoning accuracy, achieving state-of-the-art intent recognition (15% gain) and landmark identification (72% accuracy). Third, we design a reprogramming strategy with geo-semantic constraints for generating spatially plausible, personalized trajectories in data-scarce settings, showing 89% similarity to real-world data. Supported by the Kulangsu Smart Tourism System, our method provides a scalable paradigm for spatiotemporal modeling and synthetic trajectory generation, benefiting personalized services and decision-making.
- Open Access
- 12 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 362-366
Engineer Cradle: LLM-Driven Digital Twin System for Personalized Learning and Lab Management in EEE
- 12 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 362-366
Open AccessEngineer Cradle: LLM-Driven Digital Twin System for Personalized Learning and Lab Management in EEE
- Open Access
- 9 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (2) , 1-23
UVX-ray: Urban Village Safety Risk Diagnosis Leveraging Multi-Source Urban Data
Urban villages are a unique phenomenon in the downtown segments of major cities in developing countries. Most of them are heavily populated, intensely constructed, and lack infrastructure, bringing potential safety risks to their residents. Therefore, diagnosing which risk factors in urban villages contribute to the increased risk incidence is crucial for urban authorities to better renovate and manage these areas. However, traditional approaches, such as fire and traffic investigations, are labor-intensive and time-consuming, making it challenging to diagnose risks timely. To address this problem, we propose a data-driven framework that leverages heterogeneous urban data to diagnose urban village safety risks through risk-level prediction and risk factor analysis. First, we propose a crowdsensing-based approach to discover urban village potential risk hotspots and then collect contextual data from multiple sources to represent them comprehensively. Second, we propose a multi-modal representation paradigm of urban village potential risk hotspots in a multi-view manner that utilizes pre-trained models for feature extraction to effectively retain information about risk events. Finally, we design an explainable risk diagnosing model that not only predicts the risk level but also automatically highlights salient features (e.g., overcrowded restaurants for high fire risk level). Experiments using real-world data collected from 125 urban villages in Xiamen show that our approach predicts the fire risk level and the traffic risk level with 89.9% and 89.4% accuracy, respectively. Moreover, relevant risk factors in urban villages can be automatically identified for in-depth analysis by our approach.
- 9 June 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 9 (2) , 1-23
Open AccessUVX-ray: Urban Village Safety Risk Diagnosis Leveraging Multi-Source Urban Data
Urban villages are a unique phenomenon in the downtown segments of major cities in developing countries. Most of them are heavily populated, intensely constructed, and lack infrastructure, bringing potential safety risks to their residents. Therefore, diagnosing which risk factors in urban villages contribute to the increased risk incidence is crucial for urban authorities to better renovate and manage these areas. However, traditional approaches, such as fire and traffic investigations, are labor-intensive and time-consuming, making it challenging to diagnose risks timely. To address this problem, we propose a data-driven framework that leverages heterogeneous urban data to diagnose urban village safety risks through risk-level prediction and risk factor analysis. First, we propose a crowdsensing-based approach to discover urban village potential risk hotspots and then collect contextual data from multiple sources to represent them comprehensively. Second, we propose a multi-modal representation paradigm of urban village potential risk hotspots in a multi-view manner that utilizes pre-trained models for feature extraction to effectively retain information about risk events. Finally, we design an explainable risk diagnosing model that not only predicts the risk level but also automatically highlights salient features (e.g., overcrowded restaurants for high fire risk level). Experiments using real-world data collected from 125 urban villages in Xiamen show that our approach predicts the fire risk level and the traffic risk level with 89.9% and 89.4% accuracy, respectively. Moreover, relevant risk factors in urban villages can be automatically identified for in-depth analysis by our approach.
- Open Access
- 13 May 2024
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 8 (2) , 1-27
CrowdBot: An Open-Environment Robot Management System for On-Campus Services
In contemporary campus environments, the provision of timely and efficient services is increasingly challenging due to limitations in accessibility and the complexity and openness of the environment. Existing service robots, while operational, often struggle with adaptability and dynamic task management, leading to inefficiencies. To overcome these limitations, we introduce CrowdBot, a robot management system that enhances service in campus environments. Our system leverages a hierarchical reinforcement learning-based cloud-edge hybrid scheduling framework (REDIS), for efficient online streaming task assignment and dynamic action scheduling. To verify the REDIS framework, we have developed a digital twin simulation platform, which integrates large language models and hot-swapping technology. This facilitates seamless human-robot interaction, efficient task allocation, and cost-effective execution through the reuse of robot equipment. Our comprehensive simulations corroborate the system's remarkable efficacy, demonstrating significant improvements with a 24.46% reduction in task completion times, a 9.37% decrease in travel distances, and up to a 3% savings in power usage. Additionally, the system achieves a 7.95% increase in the number of tasks completed and a 9.49% reduction in response time. Real-world case studies further affirm CrowdBot's capability to adeptly execute tasks and judiciously recycle resources, thereby offering a smart and viable solution for the streamlined management of campus services.
- 13 May 2024
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 8 (2) , 1-27
Open AccessCrowdBot: An Open-Environment Robot Management System for On-Campus Services
In contemporary campus environments, the provision of timely and efficient services is increasingly challenging due to limitations in accessibility and the complexity and openness of the environment. Existing service robots, while operational, often struggle with adaptability and dynamic task management, leading to inefficiencies. To overcome these limitations, we introduce CrowdBot, a robot management system that enhances service in campus environments. Our system leverages a hierarchical reinforcement learning-based cloud-edge hybrid scheduling framework (REDIS), for efficient online streaming task assignment and dynamic action scheduling. To verify the REDIS framework, we have developed a digital twin simulation platform, which integrates large language models and hot-swapping technology. This facilitates seamless human-robot interaction, efficient task allocation, and cost-effective execution through the reuse of robot equipment. Our comprehensive simulations corroborate the system's remarkable efficacy, demonstrating significant improvements with a 24.46% reduction in task completion times, a 9.37% decrease in travel distances, and up to a 3% savings in power usage. Additionally, the system achieves a 7.95% increase in the number of tasks completed and a 9.49% reduction in response time. Real-world case studies further affirm CrowdBot's capability to adeptly execute tasks and judiciously recycle resources, thereby offering a smart and viable solution for the streamlined management of campus services.
- Open Access
- 8 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 162-164
Active 3D Mapping Leveraging Heterogeneous Crowd Robot based-on Reinforcement Learning
The paper demonstrates that the introduction of heterogeneous crowd robots improves the accuracy and completeness of indoor 3D mapping. First, three ground robots (sweeping robot, inspection robot, and guidance robot) are used to perform active exploration and mapping tasks in the iGibson indoor environment. However, the mapping results of the ground robot reveal that there are obvious blind spots in key areas (such as tables, stoves, and beds), resulting in the lack of point cloud data. To overcome this challenge, we introduce a drone for active 3D mapping using its bird’s eye view and powerful perception capabilities. Experimental results show that by introducing drones, we have successfully eliminated the blind areas of vision existing in-ground robot mapping and achieved more comprehensive and accurate mapping results. This demonstration fully demonstrates the advantages of introducing heterogeneous crowd robots, and how the complementary capabilities of different types of robots can work together to improve the indoor 3D mapping process.
- 8 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 162-164
Open AccessActive 3D Mapping Leveraging Heterogeneous Crowd Robot based-on Reinforcement Learning
The paper demonstrates that the introduction of heterogeneous crowd robots improves the accuracy and completeness of indoor 3D mapping. First, three ground robots (sweeping robot, inspection robot, and guidance robot) are used to perform active exploration and mapping tasks in the iGibson indoor environment. However, the mapping results of the ground robot reveal that there are obvious blind spots in key areas (such as tables, stoves, and beds), resulting in the lack of point cloud data. To overcome this challenge, we introduce a drone for active 3D mapping using its bird’s eye view and powerful perception capabilities. Experimental results show that by introducing drones, we have successfully eliminated the blind areas of vision existing in-ground robot mapping and achieved more comprehensive and accurate mapping results. This demonstration fully demonstrates the advantages of introducing heterogeneous crowd robots, and how the complementary capabilities of different types of robots can work together to improve the indoor 3D mapping process.