HERMAS: A Human Mobility Embedding Framework with Large-scale Cellular Signaling Data
summary
HERMAS learns city-scale mobility embeddings from a 30-day cellular signaling dataset with 130,612 users and 2,369,267 trajectories, using a two-stage tower embedding (skip-gram context + attention-modeled neighbor tower embedding to handle the Ping Pong Effect) and an RNN sequence encoder for trajectories. Results show HR@10 rising from 15.2% to 39.2%, user-profiling F1 improving by about 9%, and computation speeding up over 30×, demonstrating a practical, scalable pipeline that actually improves accuracy and efficiency.
AI-written summary · about
chapters (5)
related talks
06mCellPred: A Behavior-aware Scheme for Cellular Data Usage Prediction
06mPMF: A Privacy-preserving Human Mobility Prediction Framework via Federated Learning
07mDecentralized Attention-based Personalized Human Mobility Prediction
10mLearning Network-Based Multi-Modal Mobile User Interface Embeddings
06mExtending Coverage of Stationary Sensing Systems with Mobile Sensing Systems for Human Mobility ...
15mAccelerating Neural Recommendation Training with Embedding Scheduling
31mWi‑Fi/GSM Geolocation in Large Cities Using ML for 30 Million Users
40mSmart cars of tomorrow: real-time driving patterns