Pretrained interpretable features improve explainable generative recommendations
The Vocabulary Is Already Here: Adapting Pretrained Interpretable Features for Generative Recommendation
Information RetrievalArtificial Intelligence
Summary
Predicting what item a person might want next is hard when the system has to understand many possible items well. This paper shows that certain pretrained features, which already have clear human meanings, can be reused to represent items semantically without extra training. To do this, the authors designed a method that helps the system learn user preferences while keeping item differences clear and explains why it chooses certain items. Their approach works well and makes recommendation systems more understandable and efficient.
What this means in practice
- •For ecommerce platform teams: Build recommendation engines that use pretrained semantic features for efficient and explainable item suggestions without extra vocabulary training.$Commercial implications: Enables selling improved recommender systems that offer interpretable and efficient predictions to online retailers for better user engagement.
- •For digital media streaming engineers: Improve next-item content recommendations by leveraging pretrained interpretable features to maintain item distinctiveness with low token budgets.
Authors
Ziluowen Luo, Ruochen Liu, Jun Yin, Hao Yan, Zhengxuan Hu, Xinze Shi, Chaozhuo Li, Chengqi Zhang, Senzhang Wang
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.