WhisperRec: Latent Reasoning for Efficient Foundation Recommendation Models

2026-07-29

AI Summary

Existing large language model (LLM)-based recommender systems rely on explicit chain-of-thought (CoT) reasoning, suffering from high latency, rigid prompting templates, and limited adaptability to dynamic user interests. This work proposes WhisperRec, which introduces a novel "Latent-Reason-then-Answer" paradigm by compressing teacher-generated CoT rationales into learnable latent reasoning tokens, enabling efficient personalized inference in a compact latent space. The approach integrates multi-view adaptive chain-of-thought (MV-ACoT), a three-stage latent alignment training strategy, and a curriculum-based post-training mechanism. Experiments demonstrate that WhisperRec significantly outperforms strong baselines on both the industrial-scale Kuaishou dataset and the LLM-Rec benchmark, achieving relative improvements of 17.44% and 9.33% in SID@64, respectively, while increasing online inference throughput by over tenfold.

Abstract

Large language models (LLMs) have demonstrated strong reasoning capabilities, motivating their adoption as backbones for foundation recommendation models (FRMs). Existing approaches typically enhance recommendation with explicit Chain-of-Thought (CoT) under the Think-then-Answer paradigm. However, generating lengthy rationales introduces substantial inference overhead, while fixed CoT templates struggle to model diverse, dynamic, and context-dependent user interests. We propose WhisperRec, an efficient latent reasoning framework for FRMs. WhisperRec compresses teacher-generated CoT into learnable latent reasoning tokens, enabling a Latent-Reason-then-Answer paradigm that performs reasoning in latent space without producing verbose rationales. This design retains decision-relevant reasoning information while avoiding the latency bottleneck of autoregressive rationale generation. Specifically, it first introduces Multi-View Adaptive CoT (MV-ACoT) to construct diverse, high-quality supervision from complementary perspectives on user interests. MV-ACoT also adapts reasoning complexity to each instance, applying lightweight analysis to clear cases and targeted multi-factor reasoning to challenging ones. Building on a pre-trained FRM, WhisperRec then employs a three-stage Latent Reasoning Alignment procedure to progressively internalize teacher CoT into latent representations. Finally, curriculum-based post-training activates latent-token reasoning for downstream recommendation while preserving standard recommendation capability. Experiments on an industrial-scale Kuaishou dataset and the public Kuaishou LLM-Rec benchmark show that WhisperRec consistently outperforms explicit-CoT methods and conventional baselines. Compared with explicit CoT Think and No-Think variants, WhisperRec improves SID@64 by 17.44% and 9.33%, respectively, and achieves over 10x higher online inference throughput.

Problem

• foundation recommendation models
• Chain-of-Thought
• reasoning efficiency
• user interests
• inference overhead

Innovation

• Latent Reasoning
• Chain-of-Thought Compression
• Foundation Recommendation Models
• Multi-View Adaptive CoT
• Efficient Inference