EvoReason: Self-Evolving Reasoning Primitive-Guided On-Policy Distillation for Latent Reasoning in Generative Recommendation

Generative recommendation benefits from reasoning-enhanced inference, and latent reasoning offers an efficient paradigm by encoding intermediate reasoning processes into compact continuous representations for latency-sensitive deployment. Despite its efficiency, existing latent reasoning approaches typically rely on directly distilling raw chain-of-thought (CoT) trajectories into latent representations, assuming that textual reasoning traces provide sufficient supervision. However, recommendation reasoning trajectories contain diverse reasoning processes with redundant expressions and unstable reasoning paths, making raw CoT supervision suboptimal for learning transferable latent reasoning representations. To address this challenge, we propose EvoReason, a self-evolving latent reasoning framework that adaptively aligns explicit reasoning supervision with the student's latent reasoning space through primitive-guided on-policy distillation. First, EvoReason extracts reusable reasoning primitives from high-quality agentic recommendation trajectories, where each primitive captures an essential reasoning behavior and serves as a pseudo-tool for structured teacher reasoning. Then, based on these primitives, we equip the teacher with primitive-aware reasoning capabilities, enabling it to generate structured CoT supervision with reduced redundancy and improved consistency. Finally, during latent reasoning optimization, EvoReason introduces a self-evolving on-policy distillation mechanism, where the primitive-guided reasoning process evolves according to the student's latent reasoning outcomes. Through this closed-loop co-evolution, policy updates continuously improve latent reasoning behaviors is refined according to the resulting latent reasoning outcomes, enabling progressively better-aligned CoT supervision and more effective reasoning transfer.

Think2Go: Generative Next POI Recommendation with LLM Reasoning

Next Point-of-Interest (POI) recommendation task focuses on mining user behavioral preference patterns from historical check-ins to provide personalized suggestions for the next destination. Existing methods primarily rely on shallow contextual information and handcrafted feature interactions to predict the next POI. However, the inherent sparsity and complexity of user mobility patterns limit the computational capacity of non-reasoning models to capture deep intent, while large language models (LLMs) perform suboptimally because they lack a deep understanding of semantic IDs (SIDs) when SIDs are trained separately. To address these limitations, we propose Think2Go, a novel generative next POI recommendation framework, which enhances the model's comprehension of SID representations and explores diverse spatial-temporal patterns via test-time computational scaling. We unify supervised fine-tuning (SFT) and reinforcement learning (RL)-based reasoning within a single architecture, enabling joint optimization of memorization and adaptive reasoning to better retain user behavior patterns while exploring diverse user preferences. To further calibrate policy optimization in adaptive reasoning, we propose two advantage weighting mechanisms that integrate (1) prompt epistemic uncertainty, estimated via kernel density methods to assess the spatial-temporal periodic pattern alignment between queries and user history, promoting increased exploration under high epistemic uncertainty; and (2) reward-informed advantage scaling, captured by normalizing rewards against their maxima to adapt update magnitudes, thereby improving training stability and mitigating overfitting to noisy signals. This joint calibration forms an implicit curriculum learning strategy, delivering fine-grained, instance-aware policy updates that prevent entropy collapse and support robust exploration.

OneReason Technical Report

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style ``think before answer'' paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

SinkRec: Mitigating Semantic State Sink in Long Sequence Recommendation with Memory-Conditioned Gated Delta Networks

Linear attention provides an efficient backbone for long-sequence recommendation by avoiding the quadratic cost of standard Transformers, but its compressed recurrent state can be dominated by repetitive behavior patterns. We identify this phenomenon as semantic state sink, where recurring semantics over-occupy the recurrent state and bias subsequent readouts. To mitigate semantic state sink, we propose SinkRec, a hybrid memory-transition looped architecture that decouples collaborative behavioral pattern storage from dynamic transition modeling. SinkRec externalizes recurring local patterns into a learnable conditional memory through residual vector quantization, reinjects the retrieved codes, and exposes memory key-value pairs to the attention block. It further introduces Temporal-Aware State-Relation Differential Gated DeltaNet (TDGD), which uses memory to purify recurrent writing and reading by suppressing memory-covered updates and removing memory-aligned readout responses. This design turns recurring semantics from state-competing signals into memory-retrievable patterns, allowing the recurrent state to focus on dynamic transitions and alleviating semantic state sink with linear-time efficiency. Experiments on public and industrial datasets demonstrate the effectiveness and efficiency of SinkRec.