![[2602.21677] Trie-Aware Transformers for Generative Recommendation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21677] Trie-Aware Transformers for Generative Recommendation
Summary
The paper introduces TrieRec, a trie-aware generative recommendation method that enhances Transformers by incorporating structural inductive biases, leading to improved next-item predictions in recommendation systems.
Why It Matters
As generative AI continues to evolve, effective recommendation systems are crucial for user engagement. TrieRec's innovative approach addresses limitations in traditional models, potentially transforming how recommendations are generated and improving user experiences across various platforms.
Key Takeaways
- TrieRec improves next-item predictions by utilizing a trie structure.
- Two types of positional encodings enhance the model's understanding of item relationships.
- The method is model-agnostic and hyperparameter-free, simplifying implementation.
- Experiments show an average improvement of 8.83% across four datasets.
- This approach may set new standards for generative recommendation systems.
Computer Science > Information Retrieval arXiv:2602.21677 (cs) [Submitted on 25 Feb 2026] Title:Trie-Aware Transformers for Generative Recommendation Authors:Zhenxiang Xu, Jiawei Chen, Sirui Chen, Yong He, Jieyu Yang, Chuan Yuan, Ke Ding, Can Wang View a PDF of the paper titled Trie-Aware Transformers for Generative Recommendation, by Zhenxiang Xu and 6 other authors View PDF HTML (experimental) Abstract:Generative recommendation (GR) aligns with advances in generative AI by casting next-item prediction as token-level generation rather than score-based ranking. Most GR methods adopt a two-stage pipeline: (i) \textit{item tokenization}, which maps each item to a sequence of discrete, hierarchically organized tokens; and (ii) \textit{autoregressive generation}, which predicts the next item's tokens conditioned on the tokens of user's interaction history. Although hierarchical tokenization induces a prefix tree (trie) over items, standard autoregressive modeling with conventional Transformers often flattens item tokens into a linear stream and overlooks the underlying topology. To address this, we propose TrieRec, a trie-aware generative recommendation method that augments Transformers with structural inductive biases via two positional encodings. First, a \textit{trie-aware absolute positional encoding} aggregates a token's (node's) local structural context (\eg depth, ancestors, and descendants) into the token representation. Second, a \textit{topology-aware relative positi...
