Privacy-Preserving Sequential Recommendation: Enhancing User Privacy with Differentially Private RNNs, GRUs, and Transformers

The question behind it

Can a recommender learn what you like without learning who you are?

The core of my doctoral work. Differential privacy adds calibrated noise so that no single person can be reverse-engineered out of a model. The catch is that noise costs accuracy, and these papers measure that trade-off and find where a recommender stays useful.

A plain-language summary of this paper is on its way here. The full text will follow wherever the publisher permits me to host it.

Cite this paper

@inproceedings{sangeetha2025privacy,
  title   = {Privacy-Preserving Sequential Recommendation: Enhancing User Privacy with Differentially Private RNNs, GRUs, and Transformers},
  author  = {S Sangeetha and R Pavithra and A Ramkumar and S Balamurali and KJ Pranesh and et al.},
  booktitle = {2025 3rd International Conference on Advancements in Electrical, Electronics},
  year    = {2025},
}

Related work in this thread