Pctx: Tokenizing Personalized Context for Generative Recommendation

2025-10-24

AI Summary

Existing generative recommendation models employ static, non-personalized tokenization strategies that assign uniform semantic IDs to items solely based on item features, ignoring user-specific intent and thus failing to produce contextually differentiated item representations across users. Method: We propose the first personalized, context-aware tokenization method for generative recommendation. It dynamically constructs a learnable tokenizer from each user’s historical interaction sequence, mapping items into user-specific semantic IDs and enabling dynamic semantic encoding with multi-faceted interpretability. The method is seamlessly integrated into an autoregressive generative framework, preserving both discrete ID modeling and sequential generation capabilities. Contribution/Results: Evaluated on three public benchmarks, our approach achieves up to 11.44% improvement in NDCG@10 over non-personalized baselines. It is the first to incorporate personalized contextual information into the tokenization stage of generative recommendation, establishing a new paradigm for user-adaptive semantic representation learning.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Personalized, context-aware and across-device search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Generative recommendation (GR) models tokenize each action into a few discrete tokens (called semantic IDs) and autoregressively generate the next tokens as predictions, showing advantages such as memory efficiency, scalability, and the potential to unify retrieval and ranking. Despite these benefits, existing tokenization methods are static and non-personalized. They typically derive semantic IDs solely from item features, assuming a universal item similarity that overlooks user-specific perspectives. However, under the autoregressive paradigm, semantic IDs with the same prefixes always receive similar probabilities, so a single fixed mapping implicitly enforces a universal item similarity standard across all users. In practice, the same item may be interpreted differently depending on user intentions and preferences. To address this issue, we propose a personalized context-aware tokenizer that incorporates a user's historical interactions when generating semantic IDs. This design allows the same item to be tokenized into different semantic IDs under different user contexts, enabling GR models to capture multiple interpretive standards and produce more personalized predictions. Experiments on three public datasets demonstrate up to 11.44% improvement in NDCG@10 over non-personalized action tokenization baselines. Our code is available at https://github.com/YoungZ365/Pctx.

Problem

• Addresses static tokenization lacking personalization in generative recommendation
• Enables dynamic semantic IDs based on user context and preferences
• Improves recommendation personalization through contextual action tokenization

Innovation

• Personalized tokenizer incorporates user historical interactions
• Dynamic semantic IDs adapt to different user contexts
• Enables multiple item interpretation standards for personalization