Identifying IP Usage Scenarios: Problems, Data, and Benchmarks

Given an IP address, understanding its initial assignment,and predicting its potential usage can significantly help toward improving efficiency of many practical IP-based applications, such as Ad-recommendation, point of interest selection, fraud detection, and Internet service optimization. However, there are only a few prior studies conducted on predicting IP usage from its assignment. Surprisingly, there does not exist a benchmark dataset available to academia that can be used to investigate and develop IP usage predictions for different IP applications. In this work, we formulate the IP usage prediction problem; specifically, we collect large-scale, real-world data and extract salient features using sophisticated networking tools such as different network signals, trace route delay, IP block usage, and geographical landmarks. We showcase a series of tabular information retrieval methods to learn network signals and their interactions, and identify the IP usage scenarios. We believe our datasets and algorithms can benefit the community to facilitate relevant research in this domain, yielding more efficient and effective solutions to multiple categories of applications.