Telco Big Data Analytics for City-Scale Localization

In this project, we analyze telco big data and deploy a context-aware coarse-to-fine regression (CCR) model in Spark/Hadoop-based telco big data platform for city-scale localization. First, we design map-matching and interpolation algorithms to encode contextual information of road networks. Second, we build a two-layer regression model to capture coarse-to-fine contextual features in a short time window for improved localization performance. In our experiments, we collect 108 GPS-associated MR records in the centroid of Shanghai city with 12 × 11 square kilometers for 30 days, and measure four important properties of real-world MR data related to localization errors: stability, sensitivity, uncertainty and missing values.

This work appears in CIKM 2016.