This document proposes a framework for user identification across both online and offline datasets. The framework consists of three steps: 1) Using clustering to map IP addresses to physical locations, 2) Developing a pairwise index to reduce space and time costs for computing co-occurrences, and 3) Applying a learning-to-rank method to merge effects of multiple features from the first two steps. The goal is to build connections between online and offline data types to identify users across different datasets. Experiments were designed to evaluate the efficiency and performance of the proposed framework.