DuoLM — Video Games

Paper · GitHub · Usage guide

DuoLM brings language understanding to personalized recommendation. It combines an item encoder with a personalized ranker, using item content, user history, and feedback to model individual preferences. Item representations are computed offline and reused across recommendation requests, making inference more efficient.

This repository provides the Video Games model from DuoLM: Fast and Effective Personalized Recommendation with Language Models.

Use the model

Download the checkpoint:

hf download hulehule/DuoLM-Video-Games --local-dir checkpoints/duolm-games

Follow the usage guide to load the model and rank items. The GitHub repository includes data preparation, training, evaluation, and efficient serving.

Citation

BibTeX

@misc{duolm2026,
  title         = {{DuoLM}: Fast and Effective Personalized Recommendation with Language Models},
  author        = {Jiang Wu and Yutong Song and Hongbo Wang and Guanghe Cao and Jeff Lei Liang},
  year          = {2026},
  eprint        = {2610.24481},
  archivePrefix = {arXiv},
  primaryClass  = {cs.IR},
  doi           = {10.48550/arXiv.2610.24481},
  url           = {https://arxiv.org/abs/2610.24481}
}

License

Built on Qwen3. Model weights are released under Apache-2.0; the DuoLM code is MIT-licensed.