Street-Level Geolocation Based on Router Multilevel Partitioning

Abstract

The high-precision geolocation of Internet hosts plays an important role in many applications, such as online advertising and deception detection. The existing typical high-precision geolocation algorithms usually utilize single-hop or relative delay to geolocate an Internet host at street-level granularity. However, it is difficult to accurately measure the single-hop or relative delay within a city. This challenge sometimes results in large geolocation errors. To solve this problem, a street-level geolocation algorithm based on router multilevel partitioning is proposed. Unlike existing typical algorithms, the proposed algorithm makes a credible hypothesis that each router has a relatively stable service object for a period of time. By analyzing the connection between routers and landmarks, the possible geographic service ranges of routers are inferred from the geographic distribution of landmarks. Then, distance constraints arising from routers' service ranges are formed to estimate the geographic location of the target IP. Theoretical analysis of the geolocation error shows that the maximum and average errors of the proposed algorithm are less than those of existing typical algorithms. The proposed algorithm is evaluated by geolocating a total of 12,152 target IP addresses located in four cities in different regions. The experimental results show that, compared with the existing typical street-level geolocation algorithms SLG and NC-Geo, the average median error of the proposed algorithm decreases from 4.735 km and 3.776 km to 3.25 km, representing error reductions of approximately 31.36% and 13.96%, respectively.