An urban patrol case urgency classification method based on self-supervised clustering of a graph neural network
- 1School of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China
- 2School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou, China
Abstract
Accurate assessment of urban patrol cases plays an important role in maintaining urban safety and improving the efficiency of urban governance. The hazard level of urban patrol cases can be reflected by the urgency level and prioritizing high-urgency cases can help reduce potential risks to public safety. However, it remains challenging to assess the urgency level of urban patrol cases when reliable urgency labels are limited or unavailable. To address this problem, this paper proposes an urban patrol case urgency classification method based on self-supervised clustering of a graph neural network. First, a case-relationship graph is constructed from urban patrol cases. A graph attention-based autoencoder is then pretrained to learn node representations, and K-means is used to initialize cluster centers. Finally, representation learning and clustering are jointly optimized to identify urgency-oriented clusters without using manual labels during training. Experimental validation was conducted using 6126 urban patrol case records from Zhengzhou, China. The proposed method achieved a Recall of 0.7458, Precision of 0.7465, ACC of 0.7424, F1 of 0.7403, and NMI of 0.4517. Compared with TADW, the strongest baseline, the proposed method improved these metrics by 0.0617, 0.0479, 0.0650, 0.0690, and 0.0407, respectively. These results demonstrate that the proposed method can effectively learn urgency-related latent representations and improve the classification quality of urban patrol cases.
Related
- Concurrent Supervised-Unsupervised Representative Subspace Clustering (CSU-RSC)
- Fish 3D Clustering App (F3DCA): A Non-Supervised Clustering Method with a Machine Learning Autoencoder for Analyzing 3D Trajectories of Three Fish Species.
- Semi-supervised clustering with knowledge-guided representation learning in cryo-electron tomography
- A hybrid supervised learning model for the classification of a low-volume EEG dataset employing powerful ensembles
- Acceptability of DBS for psychiatric disorders by French psychiatrists: a network clustering analysis based on a mixed method study
BibTeX
@article{jiang2026urban,
title = {{An urban patrol case urgency classification method based on self-supervised clustering of a graph neural network}},
author = {Yirui Jiang and Hao Wu and Tiejun Yang and Jiayu Sun and Hongwei Li and Wenjie Zhu},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0360432},
url = {https://doi.org/10.1371/journal.pone.0360432},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - An urban patrol case urgency classification method based on self-supervised clustering of a graph neural network
AU - Yirui Jiang
AU - Hao Wu
AU - Tiejun Yang
AU - Jiayu Sun
AU - Hongwei Li
AU - Wenjie Zhu
PY - 2026
DA - 2026/10/09
JO - PLOS One
DO - 10.1371/journal.pone.0360432
UR - https://doi.org/10.1371/journal.pone.0360432
AB - Accurate assessment of urban patrol cases plays an important role in maintaining urban safety and improving the efficiency of urban governance. The hazard level of urban patrol cases can be reflected by the urgency level and prioritizing high-urgency cases can help reduce potential risks to public safety. However, it remains challenging to assess the urgency level of urban patrol cases when reliable urgency labels are limited or unavailable. To address this problem, this paper proposes an urban patrol case urgency classification method based on self-supervised clustering of a graph neural network. First, a case-relationship graph is constructed from urban patrol cases. A graph attention-based autoencoder is then pretrained to learn node representations, and K-means is used to initialize cluster centers. Finally, representation learning and clustering are jointly optimized to identify urgency-oriented clusters without using manual labels during training. Experimental validation was conducted using 6126 urban patrol case records from Zhengzhou, China. The proposed method achieved a Recall of 0.7458, Precision of 0.7465, ACC of 0.7424, F1 of 0.7403, and NMI of 0.4517. Compared with TADW, the strongest baseline, the proposed method improved these metrics by 0.0617, 0.0479, 0.0650, 0.0690, and 0.0407, respectively. These results demonstrate that the proposed method can effectively learn urgency-related latent representations and improve the classification quality of urban patrol cases.
SN - 1932-6203
ER -