Enhanced Dynamics of IP Allocation: Fine-Grained IP Geolocation via Temporal–Spatial Correlation

Yiyang Huang, Gang Xiong, Gaopeng Gou, Chengshang Hou, Jing Yu, Xiyuan Zhang, Xinjie Lin

Abstract: With the rapid advancement of the Internet, IP geolocation has become an indispensable part of optimizing network architecture and services. However, our analysis of millions of IP addresses reveals a significant trend: the level of dynamics in IP allocation regions enhances over time, leading to traditional static prediction strategies being ineffective and reducing geolocation performance. To address this challenge, we introduce a dynamic strategy and propose TSG, a novel fine-grained IP geolocation method based on temporal-spatial correlation. Our method employs spatial representation enhancement and temporal-spatial context graph embedding to capture the dynamics of IP allocation regions for short-term location predictions. Extensive experiments validate that TSG significantly improves the accuracy and availability of geolocation in strong dynamic scenarios. Compared to TrustGeo, the best-performing graph learning model on the Guangdong dataset, TSG reduces the median error by 72% to 1.682 km and increases the coverage rate within 5 km by 47% to 92%. Due to the TSG method not requiring additional probing, it reduces the network probing burden during large-scale updates. Additionally, this research has developed a quantitative temporal-spatial model to analyze the dynamic characteristics of IP allocation areas, offering a multi-dimensional understanding of IP spatial dynamics. This model aims to assist commercial databases in prioritizing updates for IPs with strong dynamics, and minimizes unnecessary probing. Our method significantly enhances the accuracy and availability of geolocation in dynamic scenarios, demonstrating its high efficiency and robust applicability.

External IDs:doi:10.1109/ton.2025.3603486