aiqbench analysis
This paper proposes reusing the feature indices of pretrained sparse autoencoders over language-model representations as a semantic ID vocabulary for generative recommendation, which avoids training a new tokenizer. The reported results are "competitive" rather than state-of-the-art, so the practical value depends on whether the approach holds up across other datasets and decoding setups. Its main contribution for ML readers is a partial answer to explainability: each recommendation score can be traced to human-readable feature contributions.
The Vocabulary Is Already Here: Adapting Pretrained Interpretable Features for Generative Recommendation
arXiv:2610.36763v1 Announce Type: cross Abstract: Generative recommendation (GR) predicts the next item through its semantic ID (SID). Existing methods commonly fit SID vocabularies to pretrained language model (PLM) representations of item descriptions, incurring additional training costs and making vocabulary coverage dependent on the fitting domains. We observe that pretrained sparse autoencoders expose PLM features with human-interpretable semantic annotations, allowing feature indices to serve as a reusable SID vocabulary. However, reusing this vocabulary requires preserving item distinctions under compact token budgets and learning to favor target items over candidates with shared dominant features. It also requires extending feature interpretability from item content to retrieval decisions. To address these challenges, we propose API, a framework that Adapts the Pretrained Interpretable vocabulary for effective and explainable GR without additional vocabulary learning. Within API, Pretrained Interpretable IDs (PIDs) retain selected feature indices and relative activation strengths for item discrimination. With PIDs as supervision, the Semantic Prediction Adapter (SPA) learns user preferences from interaction histories through joint feature-index and strength prediction. SPA combines both predictions to score candidate items, with item-level supervision encouraging the observed next item to outrank its alternatives. API further explains candidate score differences through feature contributions paired with their pretrained semantic annotations. Extensive experiments show that API achieves competitive performance across backbones and decoding schemes, while preserving item distinctions under compact token budgets and enabling explainable retrieval.
Abstract:Generative recommendation (GR) predicts the next item through its semantic ID (SID). Existing methods commonly fit SID vocabularies to pretrained language model (PLM) representations of item descriptions, incurring additional training costs and making vocabulary coverage dependent on the fitting domains. We observe that pretrained sparse autoencoders expose PLM features with human-interpretable semantic annotations, allowing feature indices to serve as a reusable SID vocabulary. However, reusing this vocabulary requires preserving item distinctions under compact token budgets and learning to favor target items over candidates with shared dominant features. It also requires extending feature interpretability from item content to retrieval decisions. To address these challenges, we propose API, a framework that Adapts the Pretrained Interpretable vocabulary for effective and explainable GR without additional vocabulary learning. Within API, Pretrained Interpretable IDs (PIDs) retain selected feature indices and relative activation strengths for item discrimination. With PIDs as supervision, the Semantic Prediction Adapter (SPA) learns user preferences from interaction histories through joint feature-index and strength prediction. SPA combines both predictions to score candidate items, with item-level supervision encouraging the observed next item to outrank its alternatives. API further explains candidate score differences through feature contributions paired with their pretrained semantic annotations. Extensive experiments show that API achieves competitive performance across backbones and decoding schemes, while preserving item distinctions under compact token budgets and enabling explainable retrieval.
Information Retrieval (cs.IR); Artificial Intelligence (cs.AI)
arXiv:2610.36763 [cs.IR]
(or arXiv:2610.36763v1 [cs.IR] for this version)
https://doi.org/10.48550/arXiv.2610.36763 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 318 words
Related coverage
Evidence Before Sampling: Interpretable Implicit Negative Candidate Discovery for Recommendation
arXiv · Oct 7, 2026
MediRec: Enhancing Chinese Medication Recommendation with Explainable Clinical Reasoning
arXiv · Aug 6, 2026
SemPOI-RL: Aligning LLM Semantic Reasoning for Interpretable Out-of-Town POI Sequential Generation
arXiv · Sep 1, 2026
Discovering Topics from Linear Autoencoders
arXiv · Oct 27, 2026