Enhancing Kepler.gl for processing Google Maps Timeline data

Tianqi Xia, Hiroshi Kanasugi, Zipei Fan, Ruochen Si, Chuyao Feng, Hiroshi Takayasu, Ryosuke ShibasakiAbstracts 2021 · DOI

Introduction

Human mobility data is a critical asset for urban planning and social science research. However, researchers often face a dilemma: large-scale mobile carrier data is frequently inaccessible or too coarse, while small-scale studies relying on manual location logging are prone to user error and "tracking fatigue."

Google Maps Timeline (GMT) data presents a viable middle ground, offering high-quality, automatically tracked location history with rich semantic detail (e.g., identifying a trip as "commuting via train"). Despite its utility, GMT data is highly sensitive, revealing precise home addresses and daily routines. To address this, researchers have developed an enhanced version of the open-source tool Kepler.gl, specifically designed to process and visualize GMT data while prioritizing participant privacy through client-side desensitization and advanced filtering.

Methodology

The proposed system extends the core functionality of Kepler.gl in three primary areas: file reading, advanced filtering, and privacy-preserving desensitization.

File Reading and Visualization

Standard Kepler.gl does not natively support GMT formats. The enhanced system converts raw GMT files into standard geospatial layers. GPS points are transformed into "point" or "trip" layers, while semantic data—such as travel between two points—is converted into "arc" or "line" layers for intuitive visualization.

Privacy-Preserving Desensitization

A core feature of the system is its "front-end" privacy shield. By performing data processing and filtering within the user's browser, sensitive raw data never needs to be uploaded to a central server. The system employs specific desensitization methods to generalize data, preserving its research value while stripping away identifying details.

Desensitization methods for raw and semantic location history

As shown in the framework above, raw coordinates are converted to mesh grids, and specific timestamps are indexed. For semantic data, the system generalizes addresses and converts specific facility names (e.g., a specific store name) into general facility types (e.g., "Cafe").

System Enhancements and Results

The enhanced Kepler.gl provides several functional improvements over the standard version, particularly in its ability to handle complex spatial and temporal logic natively.

Comparison of features between standard Kepler.gl and the enhanced system

Key results of the implementation include:

  • Native GMT Support: The system can read GMT data as tables and visualize it across multiple layers automatically.
  • Advanced Spatial and Temporal Logic: It supports filtering data using regions defined in external GeoJSON files and offers granular temporal filters (e.g., weekday vs. weekend).
  • Automated Data Processing: The system includes tools for table joins, data aggregation based on region geometry, and POI (Point of Interest) conversion, which are unavailable in the base version of Kepler.gl.

Discussion and Future Work

While the system significantly lowers privacy risks by keeping data processing local, some limitations remain. Geocoding and complex address analysis still require server-side requests, representing a minor privacy bottleneck. Furthermore, the current version does not yet perfectly align GPS traces to road networks.

Future development will focus on implementing "map matching" to generate more accurate road-network-based mobility data and further reducing server dependencies to ensure a fully secure, client-side research environment.

Summary updated

BibTeX
@inproceedings{xia2021enhancing,
  author = {Tianqi Xia and Hiroshi Kanasugi and Zipei Fan and Ruochen Si and Chuyao Feng and Hiroshi Takayasu and Ryosuke Shibasaki},
  title = {{Enhancing Kepler.gl for processing Google Maps Timeline data}},
  booktitle = {Abstracts},
  year = {2021},
  doi = {10.5194/ica-abs-3-319-2021},
}
Figure