Changjiang Han

Ye Yuan, Rui Song, Weien Li et al.

Social deduction games have become a popular testbed for probing reasoning, deception, coordination, and belief modeling in Large Language Model (LLM) agents. However, most environments are scored only by game outcomes such as win rates and largely remain to text-only interaction, making it difficult to tell whether an agent's language is actually grounded in what it perceived and did, or to identify the failure modes underlying its behavior. To address this gap, we introduce QUACK, an open-source environment and evaluation framework for auditing the grounding of agent language in multimodal social reasoning. QUACK evaluates agents at three levels: game outcomes, behavioral trajectories, and utterance-level consistency. Its core Statement Verification Pipeline reconstructs each agent's ground-truth trajectory from engine logs and checks every discussion claim against it, automatically flagging spatial hallucination, unsupported accusation, deception collapse, and language-action inconsistency. Evaluating three frontier VLMs in both homogeneous and cross-model adversarial settings, we find that even the strongest agent hallucinates 15.1% of its verifiable spatial claims and makes over half of its accusations without grounded evidence. We release the full engine, evaluation framework, toolkit, and logs at https://github.com/AAAAA-Academia-Attractions/QUACK.

Weixu Zhang, Ye Yuan, Changjiang Han et al.

Large Language Models (LLMs) exhibit strong implicit personalization ability, yet most existing approaches treat this behavior as a black box, relying on prompt engineering or fine tuning on user data. In this work, we adopt a mechanistic interpretability perspective and hypothesize the existence of a sparse set of Preference Heads, attention heads that encode user specific stylistic and topical preferences and exert a causal influence on generation. We introduce Differential Preference Steering (DPS), a training free framework that (1) identifies Preference Heads through causal masking analysis and (2) leverages them for controllable and interpretable personalization at inference time. DPS computes a Preference Contribution Score (PCS) for each attention head, directly measuring its causal impact on user aligned outputs. During decoding, we contrast model predictions with and without Preference Heads, amplifying the difference between personalized and generic logits to selectively strengthen preference aligned continuations. Experiments on widely used personalization benchmarks across multiple LLMs demonstrate consistent gains in personalization fidelity while preserving content coherence and low computational overhead. Beyond empirical improvements, DPS provides a mechanistic explanation of where and how personalization emerges within transformer architectures. Our implementation is publicly available.

Changjiang Han, Qingyang Li, Yaqiang Zang et al.

Making LLM-based generative recommendation models stronger and more personalized through natural language and explicit reasoning is a widely anticipated yet still unsolved goal. Such models cast recommendation as autoregressively generating an item's semantic-ID (SID), a short tuple of discrete codes, so that recommending well reduces to emitting the right SID. In this setting the model verbalizes its knowledge poorly, and text and SID tokens live in misaligned embedding spaces. Deep reasoning therefore rarely turns into a correct SID, and enabling explicit "thinking" often gives no gain or even hurts. The deeper cause is that a compact SID cannot hold content and collaborative signal at once: the two compete, and collaboration loses. Because a mis-predicted SID is a wrong recommendation, this caps accuracy directly. Costly multi-round training barely helps, and few methods try to supply the missing signal at inference time. What is missing is a reliable channel that carries collaborative signal into SID generation. We therefore propose a framework, guided by personalized natural language, that adds hierarchical collaborative cues as the model generates, without altering the backbone or retraining the SIDs. Rather than mapping language onto SIDs directly, it uses language to attach analyzable links between collaborative patterns and their audiences, restoring the collaborative signal that SIDs miss. The result is consistent gains in recommendation accuracy, grounding generation in collaborative structure at inference time rather than relying on explicit reasoning or retraining.

Guo Tang, Hanye Wu, Changjiang Han et al.

Generative recommendation unifies retrieval and ranking within a single model by autoregressively decoding semantic identifier (SID) sequences. Yet reliably incorporating behavior signals from clicks, cart additions, and orders remains challenging. Existing approaches either jointly optimize generative and discriminative objectives, requiring delicate trade-offs, or use a separate ranker as a post-hoc reinforcement-learning reward, risking out-of-distribution scoring and reward misalignment. We propose OxygenREC-v2, a generative recommender that Internalizes Discrimination into Generative Recommendation (IDGR). Rather than adding a separate discriminative objective, OxygenREC-v2 uses logged behavior to condition generation and supervise training. During pre-training, a behavior instruction conditions generation on the target behavior. During post-training, future interaction behaviors are exploited as privileged knowledge in our entropy-aware trajectory optimization self-distillation framework, enabling reward-model-free policy optimization. Throughout both training stages, OxygenREC-v2 maintains a single unified backbone. We implement OxygenREC-v2 as a 3B-parameter, 1B-activated MoE and deploy it on JD.com's large-scale e-commerce platform. Across multiple online A/B tests, OxygenREC-v2 improves user click-through conversion rate (UCTCVR) by 1.6--4.4% and GMV by 2.8--6.8% over OxygenREC-v1.