Reproduce “Beyond Item Order: Temporal Gap Tokenization for Generative Recommendation with Semantic IDs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chengkai Huang, Tianqi Gao, Hongtao Huang, Quan Z. Sheng, Lina Yao
ChronoSID enhances semantic-ID based generative recommendation by integrating inter-interaction time gaps into item representation learning and the sequence generation process. This framework demonstrates superior performance against existing SID-based methods and competitive results against conventional sequential recommenders across eight Amazon datasets, improving Recall@5 from 0.0388 to 0.0417 on the Musical Instruments dataset.
Key insights
- Integrating inter-interaction time gaps explicitly into SID-based generative recommenders leads to substantial performance gains.
- Directly exposing the sequence-to-sequence generator to temporal context via interleaved gap tokens is a more impactful enhancement than solely regularizing item representations with temporal predictability.
- ChronoSID's temporal modeling proves especially effective for interactions separated by longer time gaps, indicating its value in addressing preference drift.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.