Block 1
Foundations and SASRec
Sequential recommendation, chronological data preparation, item embeddings, causal self-attention, and next-item training. SASRec and a small MovieLens example make the model’s computations inspectable.
Transformer models for recommendation
Transformer-based
Sequential Recommender Systems
From user histories to next-item predictions: an introduction to sequence models, their training objectives, and the choices that shape evaluation and serving.
Jan Malte Lichtenberg Aleksandr V. Petrov
The slides include interactive model playgrounds. For offline use,
unzip the download and open tutorial/index.html in your browser.
Programme
Block 2
Objectives, evaluation, and scale
Training objectives and BERT4Rec, evaluation choices, large item catalogs, content representations, cold start, serving, and recent directions in generative recommendation.
Audience and preparation
For researchers and practitioners familiar with basic recommendation and machine-learning concepts. No prior experience with sequential recommenders is required.
Bring a laptop to try the playgrounds. Models run in your browser; no account, software installation, or inference service is needed.
Citation
BibTeX
If you use these tutorial materials, please cite:
@inproceedings{lichtenberg2026transformer,
title={Transformer-based Sequential Recommender Systems},
author={Lichtenberg, Jan Malte and Petrov, Aleksandr V},
booktitle={Proceedings of the 20th ACM Conference on Recommender Systems},
pages={1911--1912},
year={2026}
}