Abstract
Large-scale digital recommendation systems may face challenges associated with data sparsity, which can make it difficult to provide relevant personalization for users with limited interaction history. The synthetic user interaction profiles can be generated using a generative model, such as a variational auto-encoder, to address such challenges. A variational auto-encoder may be trained on real user profiles to learn a compressed, probabilistic latent space representation of user-item interactions. New, statistically plausible user profiles can then be synthesized by sampling from this latent space and processing the samples through a decoder network. The generated synthetic profiles can be used to augment training datasets, potentially improving model performance and mitigating the cold-start problem, and can also be applied to tasks such as privacy-preserving analytics and system testing.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Chatterjee, Tamojit and Acharya, Sayan, "Generating Synthetic User Interaction Profiles via a Variational Auto-Encoder", Technical Disclosure Commons, (February 18, 2026)
https://www.tdcommons.org/dpubs_series/9340