Batch and negative sampling design for human mobility graph neural network training

Jiaxin Du, Xinyue Ye

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This study presents a deep learning approach to human mobility prediction within a network science framework. We proposed a human mobility graph that defines two types of nodes - -people and locations - -and employs domain-specific attributes to capture the temporal and spatial dynamics of human movement. The graph is too large to fit into a single GPU, necessitating the design of a specialized mini-batching strategy that incorporates behavioral aspects. The study also introduces a location-aware negative sampling method to enhance the training process. The results achieved the top 10 accuracy based on GEO-BLEU and Dynamic Time Warping in HuMob Challenge 2023. It serves as a call to action for researchers in the field to consider the entire deep learning process in human mobility studies. Looking ahead, the study advocates for the release of more comprehensive datasets to further enrich our understanding of human mobility patterns.

Original languageEnglish (US)
Title of host publicationHuMob 2023 - 1st ACM SIGSPATIAL International Workshop on the Human Mobility Prediction Challenge
EditorsTakahiro Yabe, Kota Tsubouchi, Toru Shimizu, Yoshihide Sekimoto, Kaoru Sezaki, Esteban Moro, Alex Sandy Pentland
PublisherAssociation for Computing Machinery, Inc
Pages47-50
Number of pages4
ISBN (Electronic)9798400703560
DOIs
StatePublished - Nov 13 2023
Externally publishedYes
Event1st ACM SIGSPATIAL International Workshop on the Human Mobility Prediction Challenge, HuMob 2023 - Hamburg, Germany
Duration: Nov 13 2023 → …

Publication series

NameHuMob 2023 - 1st ACM SIGSPATIAL International Workshop on the Human Mobility Prediction Challenge

Conference

Conference1st ACM SIGSPATIAL International Workshop on the Human Mobility Prediction Challenge, HuMob 2023
Country/TerritoryGermany
CityHamburg
Period11/13/23 → …

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Computer Science Applications
  • Control and Systems Engineering
  • Transportation

Keywords

  • graph neural network
  • heterogenous graph
  • human mobility
  • mini-batching
  • time series

Fingerprint

Dive into the research topics of 'Batch and negative sampling design for human mobility graph neural network training'. Together they form a unique fingerprint.

Cite this