HoloRec: Holistic Encoding and Interleaved Reasoning for Generative Recommendation

2026-06-13

AI Summary

This work addresses key limitations in generative recommendation systems—namely, flat semantic representations, insufficient hierarchical reasoning capabilities, and reliance on costly annotated chain-of-thought (CoT) data—by introducing an endogenous CoT mechanism. The proposed approach constructs a hierarchical semantic encoding matrix through multi-granularity nested residual quantization, unifying representation, reasoning, and generation within a single framework. It incorporates both non-thinking and thinking inference modes, enabling dynamic, annotation-free multi-step reasoning during generation. Combined with a holistic reconstruction loss, lightweight multi-granularity supervision alignment, and an interleaved reasoning strategy, the model significantly outperforms baseline methods across multiple public recommendation benchmarks, demonstrating particularly strong performance in data-sparse scenarios while achieving higher accuracy with modest additional inference overhead in thinking mode.

Abstract

Generative recommendation models that formulate the task as sequence generation overcome the objective fragmentation problem of traditional cascade architectures, yet existing approaches still suffer from flat semantic representations lacking hierarchical structure for multi-step reasoning and an externally constructed chain-of-thought (CoT) that requires expensive annotations and remains disconnected from the generation objective. We propose HoloRec, an endogenous chain-of-thought recommendation mechanism that unifies representation, reasoning, and generation by constructing a hierarchical semantic encoding matrix via multi-granularity nested residual quantization optimized by a holistic reconstruction loss. HoloRec supports two inference modes: a non-thinking mode that uses lightweight multi-granularity supervised alignment for fast prediction, and a thinking mode that employs an interleaved reasoning scheme to generate CoT steps on the fly, directly embedding reasoning into the generation process without external data. Experiments on multiple public recommendation datasets demonstrate that HoloRec consistently outperforms baselines, with especially significant gains in sparse scenarios, and the thinking mode achieves better accuracy than the non-thinking mode with only modest inference overhead.

Problem

• generative recommendation
• chain-of-thought
• hierarchical representation
• multi-step reasoning
• semantic encoding

Innovation

• generative recommendation
• chain-of-thought
• hierarchical semantic encoding
• nested residual quantization
• interleaved reasoning