Aims and Approach
The project "Anonymization and Synthesis of Click Paths and Behavior on the Web (SynthiClick)" aims to develop new concepts for anonymization and data synthesis that allow reach measurements, usage analyses and benchmarking in online marketing in compliance with data protection laws. For this purpose, generative machine learning methods are used to synthesise attribute-true data sets from survey and tracking data of German websites. The researchers are developing metrics to compare the anonymous, synthesised data with the survey and tracking data sets.
At the same time, they check whether there is actually no more personally identifiable information. Subsequently, the metrics are used to perform sustainable anonymisations using synthesis algorithms, which then generate freely usable data sets. The developed algorithms and generated data will be transferred to the public domain.
Who is who?
In this project, the following academic and private partners are working together:
KIT
Karlsruhe
KASTEL Security Research Labs
Chair of IT Security
Prof. Dr. Strufe
INFOnline GmbH
Bonn
Service provider for digital audience measurement
Leuphana University
Lüneburg
Institute of Information Systems
Machine Learning
Prof. Dr. Brefeld
Results from the Project
SynthiClick has yielded a broad range of technical and scientific results
A demonstrator that shows the core technical contributions can be found at this link
The project also yielded a number of scientific publications:
- Morsbach, Felix, Jan Ludwig Reubold, and Thorsten Strufe. "R+ R: Understanding Hyperparameter Effects in DP-SGD." Proceedings of the 40th Annual Computer Security Applications Conference. 2024.
- Miranda-Pascual, Àlex, et al. "An overview of proposals towards the privacy-preserving publication of trajectory data." International Journal of Information Security (2024): 1-37.
- Guerra-Balboa, Patricia, et al. "Composition in differential privacy for general granularity notions." IEEE 37th Computer Security Foundations Symposium (CSF). IEEE, 2024.
- Wohlstein, Moritz and Ulf Brefeld. “Toward Learning Distributions of Distributions” Proceedings of the Northern Lights Deep Learning Conference, 2025.
Funding
SynthiClick is a project that is funded by the European Union as part of the NextGenerationEU program.
Contact
For more information about the project and our partners, you can contact our coordinator Prof. Strufe.
thorsten.strufe@kit.edu