S$^2$GR: Stepwise Semantic-Guided Reasoning in Latent Space for Generative Recommendation

TL;DR

This work tackles the limited reasoning depth in generative recommendation by introducing S$^2$GR, which interleaves stepwise thinking tokens with SID generation to produce hierarchically grounded, interpretable reasoning paths. A collaborative and balanced RQ-VAE (CoBa RQ-VAE) leverages item co-occurrence signals and capacity-balancing losses to strengthen the coarse-to-fine semantic structure of SIDs. Stepwise reasoning tokens are supervised through coarse-grained semantic alignment, contrastive signals, and an auxiliary holistic semantic decoder, ensuring reliable and balanced computation across SID levels. Extensive offline and online evaluations demonstrate that S$^2$GR consistently outperforms state-of-the-art baselines, with notable gains on large-scale industrial data and verified business impact in online A/B tests.

Abstract

Generative Recommendation (GR) has emerged as a transformative paradigm with its end-to-end generation advantages. However, existing GR methods primarily focus on direct Semantic ID (SID) generation from interaction sequences, failing to activate deeper reasoning capabilities analogous to those in large language models and thus limiting performance potential. We identify two critical limitations in current reasoning-enhanced GR approaches: (1) Strict sequential separation between reasoning and generation steps creates imbalanced computational focus across hierarchical SID codes, degrading quality for SID codes; (2) Generated reasoning vectors lack interpretable semantics, while reasoning paths suffer from unverifiable supervision. In this paper, we propose stepwise semantic-guided reasoning in latent space (S$^2$GR), a novel reasoning enhanced GR framework. First, we establish a robust semantic foundation via codebook optimization, integrating item co-occurrence relationship to capture behavioral patterns, and load balancing and uniformity objectives that maximize codebook utilization while reinforcing coarse-to-fine semantic hierarchies. Our core innovation introduces the stepwise reasoning mechanism inserting thinking tokens before each SID generation step, where each token explicitly represents coarse-grained semantics supervised via contrastive learning against ground-truth codebook cluster distributions ensuring physically grounded reasoning paths and balanced computational focus across all SID codes. Extensive experiments demonstrate the superiority of S$^2$GR, and online A/B test confirms efficacy on large-scale industrial short video platform.

Figures (6)

• Figure 1: The illustrations of latent reasoning paradigms. The left and middle subfigures show the processes in sequential and generative recommendation, where multiple latent representations are autoregressively generated and refined before generating the final user representation or SID. The right subfigure presents our proposed stepwise reasoning paradigm for generative recommendation, wherein coarse-grained semantic categories are thinked before each SID code generation, serving as an transitional step toward the next SID code.
• Figure 2: Framework of Collaborative and Balanced RQ-VAE. We first construct an item co-occurrence graph based on users’ historical interactions, where the color intensity indicates the frequency of co-occurrence. The original item representations are then enriched with co-occurrence information via neighborhood aggregation. During residual quantization training, we introduce a within-codebook distribution uniformity constraint to prevent codebook representation collapse. Additionally, when activating codewords, we dynamically adjust the selection probability of each codeword based on historical activation records to achieve balanced codebook usage.
• Figure 3: Framework of Stepwise Semantic-Guided Reasoning. We interleave stepwise thinking tokens within the semantic ID generation process, which represent coarse-grained category semantics, guiding the subsequent SID code to be drawn from the semantic cluster associated with that thinking token. Each thinking token is supervised by the coarse-grained semantics obtained via codebook clustering, while the initial thinking token is additionally regularized to integrate the holistic item semantics, ensuring the reliability of the reasoning path.
• Figure 4: The performance variation across different cluster number of coarse-grained semantic guidance.
• Figure 5: The performance comparison under different user interaction sequence lengths on industrial dataset.