arXiv subjects

Pengjie Wang

Publications and source records attributed to Pengjie Wang.

At least 19 recordsLinked to original sources

arXiv2026

SPAR: Enhancing Industrial-Scale Generative POI Recommendation via Real-World Spatial Perception

Generative Point-of-Interest (POI) recommendation, autoregressively generating a target POI's semantic ID (SID), holds great promise for Location-Based Services, where a recommendation helps only if the user can reach it. Yet, existing methods operate within an interest space defined by behavior sequences and collaborative signals, where geography enters only as a textual attribute of the SID, leaving no explicit mechanism to learn or preserve how urban places are related by distance, direction, and reachability; their predictions are thus behaviorally plausible yet far from the user's real-time location. We argue that such services require injecting real urban spatial knowledge into the interest space, rather than inferring geography from behavior alone. Hence, we propose SPAR, a unified framework whose three synergistic stages jointly construct, cultivate, and preserve urban spatial knowledge: (1) at the tokenization level, Spatially-Intrinsic SID (SI-SID) explicitly encodes longitude--latitude coordinates into a sinusoidal geospatial embedding and fuses it with the textual semantic embedding, producing identifiers via RQ-Kmeans that are simultaneously semantically and geographically consistent; (2) at the cognition level, Multi-Granular Geospatial CPT (MG-CPT) continually pre-trains the base LLM on 25 curated geospatial datasets organized into three tiers of basic attributes, pairwise relations, and city-scale navigation, so that scattered POIs cohere into a connected urban space; and (3) at the adaptation level, Task-Vector Anchored SFT (TV-SFT) anchors the acquired spatial knowledge as a frozen parameter-space task vector to prevent its catastrophic forgetting during behavioral fine-tuning, thereby fusing the two spaces. Extensive quantitative and visualization experiments on two public and four industrial-scale datasets demonstrate the effectiveness of SPAR.

cs.IR↗

arXiv2026

HeMix: Scaling Industrial Ranking Models with Heterogeneous Token Mixing

Scaling up ranking models for industrial recommender systems faces two critical challenges: (C1) existing sequence tokenization fails to jointly capture context-aware and context-invariant user intent from heterogeneous behavior sources, and (C2) prevailing interaction mechanisms are both computationally expensive and semantically homogeneous, limiting prediction quality under strict online latency constraints. We propose \textbf{HeMix}, a scalable ranking model that unifies query-mixed sequence tokenization with heterogeneous feature interaction. To address (C1), HeMix introduces a \textit{Query-Mixed Interest Extraction} module that employs dynamic and fixed queries to simultaneously model context-aware and context-invariant interests from global and real-time behavior sequences. To address (C2), we design the \textit{HeteroMixer} block, comprising Multi-Head Token Fusion, Heterogeneous Mixed-Token Interaction and Group-Aligned Reconstruction, as an efficient alternative to self-attention that enables multi-granularity cross-feature modeling at linear cost. Crucially, HeMix scales smoothly from ${\sim}100$M to ${\sim}1500$M parameters by independently expanding block depth and token dimension, yielding steady accuracy gains without architectural redesign. Experiments on industrial-scale data show that HeMix achieves $+1.64\%$ relative CTR-AUC over the DLRM baseline at the ${\sim}100$M scale while requiring fewer GFLOPs than the strongest competitor. Deployed on the AMAP APP, HeMix yields +0.88\% GMV, +2.74\% PV\_CTR and +0.84\% UV\_CVR over the production baseline in online A/B tests.

cs.IR↗

arXiv2026

GeoGR: Enabling Spatio-Temporal Aware Industrial-scale Generative POI Recommendations

Next Point-of-Interest (POI) prediction is a fundamental task in location-based services (LBS), especially critical for large-scale navigation platforms such as AMAP that serve billions of users in diverse lifestyle scenarios. Although recent POI recommendation approaches based on SIDs have achieved promising performance, they struggle in complex, sparse real-world environments due to two key limitations: (1) inadequate modeling of high-quality SIDs that capture cross-category spatio-temporal collaborative relationships, and (2) poor alignment between large language models (LLMs) and the POI recommendation task. To this end, we propose GeoGR, a geographic generative recommendation framework tailored for navigation-based LBS like AMAP, which perceives changes in users' contextual states and enables spatio-temporal aware POI recommendation. GeoGR features a two-stage design: (i) a geo-aware SID tokenization pipeline that explicitly learns spatio-temporal collaborative semantic representations via geographically constrained co-visited POI pairs, contrastive semantic representation learning, and iterative refinement; and (ii) a multi-stage LLM training strategy that aligns non-native SID tokens through continued pre-training with multiple prompt templates and enables autoregressive POI generation via supervised fine-tuning. Extensive experiments on multiple real-world datasets demonstrate GeoGR superiority over state-of-the-art baselines. Moreover, the deployment on the large-scale AMAP platform over three months, serving millions of users and delivering significant online gains of +2.91% in WINRATE and +5.55% in PV_CTR, confirms its practical effectiveness and scalability in production.

cs.IR↗

arXiv2026

HF-SID: High-Fidelity Semantic IDs for Generative Retrieval in Location-Based Services

Generative retrieval has attracted increasing attention in Location-Based Services (LBS), where each Point-of-Interest (POI) is represented as a Semantic ID (SID). As the SID is the only channel through which POI information reaches the generative model, whatever it fails to preserve is irrecoverable at decoding time, and LBS retrieval is especially sensitive to the fine-grained differences that existing SIDs blur. Specifically, (1) LLMs embed continuous coordinates discontinuously, so their numeric differences do not reflect true geographic distance; (2) dynamic numerical attributes differ vastly in scale, so an identical gap may be decisive for one attribute yet negligible for another; and (3) short text cannot convey hierarchical affiliation, as text-similar POIs may belong to different hierarchies. We therefore propose HF-SID, which restores geographic, numerical, and structural fidelity at the representation stage, before any information is committed to a discrete code. It transforms coordinates into a continuous 3D Cartesian form and encodes each numerical value as a single unit, consolidated inside the LLM by Geo-CPT and Num-CPT with type-aware embeddings; a Structure-based Contrastive Learning objective, applied only to the last-layer residual, then separates co-located POIs that share a coarse tag but differ at the fine level. Because these mechanisms enrich the representation rather than lengthen the identifier, HF-SID uses a 3-token SID at no extra decoding cost. On a large-scale industrial

cs.IR↗

arXiv2026

Resonant Far-Infrared Spectroscopy of Flat-Band Fermions in Magic Angle Graphene

Moiré engineering in twisted two-dimensional (2D) materials radically alters low-energy bands, interactions and topological quantum states. Despite extensive studies, optical spectroscopy of interacting moiré bands in the characteristic far-infrared (FIR) regime has remained largely unexplored due to extreme experimental challenges. Using a newly developed millikelvin FIR platform, we report the observation of the long-sought-after characteristic FIR resonances of flat-band electrons in magic-angle twisted bilayer graphene (MATBG). We observe highly tunable spectroscopic signatures of interacting light and heavy fermions that constitute the flat bands in MATBG. Using the topological heavy-fermion model (THF), we show that itinerant topological electrons act as an "antenna" that couples strongly to the optical field, with resonant frequencies renormalized by the hybridization with localized heavy electrons. We establish optical selection rules of MATBG which uncovers the key symmetry governing light-heavy fermion hybridization. At charge neutrality, we observe pronounced resonances at energies below the on-site Coulomb energy, implying the emergence of new many-body modes. Our experiments and modeling provide a fundamental understanding of light-matter interactions in MATBG and enable resonant optical spectroscopy of moiré bands down to millikelvin temperatures.

cond-mat.mes-hall↗

arXiv2026

MOON2.0: Dynamic Modality-balanced Multimodal Representation Learning for E-commerce Product Understanding

Recent Multimodal Large Language Models (MLLMs) have significantly advanced e-commerce product understanding. However, they still face three challenges: (i) the modality imbalance induced by modality mixed training; (ii) underutilization of the intrinsic alignment relationships among visual and textual information within a product; and (iii) limited handling of noise in e-commerce multimodal data. To address these, we propose MOON2.0, a dynamic modality-balanced MultimOdal representation learning framework for e-commerce prOduct uNderstanding. It comprises: (1) a Modality-driven Mixture-of-Experts (MoE) that adaptively processes input samples by their modality composition, enabling Multimodal Joint Learning to mitigate the modality imbalance; (2) a Dual-level Alignment method to better leverage semantic alignment properties inside individual products; and (3) an MLLM-based Image-text Co-augmentation strategy that integrates textual enrichment with visual expansion, coupled with Dynamic Sample Filtering to improve training data quality. We further release MBE2.0, a co-augmented Multimodal representation Benchmark for E-commerce representation learning and evaluation at https://huggingface.co/datasets/ZHNie/MBE2.0. Experiments show that MOON2.0 delivers state-of-the-art zero-shot performance on MBE2.0 and multiple public datasets. Furthermore, attention-based heatmap visualization provides qualitative evidence of improved multimodal alignment of MOON2.0.

cs.CV↗

arXiv2026

AlphaOracle: Oracle bone script decipherment via human-workflow-inspired deep learning

Approximately 3,000 of the 4,500 oracle bone script (OBS) characters remain undeciphered due to fragmentary inscriptions and sparse evidence. Current AI approaches fail to replicate expert workflows that integrate form analysis, contextual semantics, and philological reasoning. We introduce AlphaOracle, a human-workflow-inspired framework that systematizes OBS decipherment using the largest digitized corpus to date. Its multi-stage pipeline comprises: (i) rubbing parsing; (ii) radical-based morphological analysis with diachronic modeling; (iii) contextual retrieval with semantic alignment; and (iv) philological validation against classical sources. Each stage generates explicit, confidence-weighted evidence chains, culminating in interpretable reports for scholarly verification. Across multiple test characters, AlphaOracle's readings strongly agreed with expert interpretations. In a study of 86 domain specialists, it reduced analysis time by 64% and 79% of participants rated it highly useful. Notably, AlphaOracle resolves the character "Lao" as a toponymic or clan designation, offering concrete revisions to Shang administrative and social interpretations. These results suggest that computational methods aligned with philological practice can facilitate OBS research and provide a conceptual reference for studies of other undeciphered scripts.

cs.HC↗

arXiv2026

DeltaV: Thinking with Visual State Updates in Unified Large Multimodal Models

Current Unified Large Multimodal Models (ULMMs) support interleaved multimodal reasoning through textual reasoning and intermediate visual states, but typically generate each visual state as a full image. This full-image generation paradigm introduces substantial visual-token redundancy and dilutes supervision on sparse yet reasoning-critical state transitions. We propose DeltaV, a ULMM that replaces full-image generation with visual updates. Conditioned on historical visual states, DeltaV incrementally predicts compact update tokens that capture the visual changes across reasoning steps, avoiding repeated modeling of unchanged content. To align the token budget of each update with the magnitude of visual change, DeltaV introduces a temporal similarity (TSIM) Router, which stops allocating tokens once the marginal reconstruction gain falls below a threshold. To support more diverse and generalizable reasoning, we further construct StructCoT, a large-scale interleaved multimodal reasoning dataset with 1.05M samples spanning 44 task domains. Experiments show that the visual-update paradigm reduces newly generated visual tokens by 55.6\% on average without compromising reconstruction fidelity, and improves multimodal reasoning by 3.3\% over full-image generation. Trained with StructCoT and large-scale multimodal data, DeltaV-2B further outperforms substantially larger open-source models by 8.4\% on in-domain multimodal reasoning evaluations and surpasses the comparable-scale Qwen3-VL-2B by 5.9\% on external multimodal reasoning and understanding benchmarks. Code, models, and StructCoT will be released at https://github.com/Pengjie-W/DeltaV.

cs.CV↗

arXiv2026

GFlowGR: Fine-tuning Generative Recommendation Frameworks with Generative Flow Networks

Generative recommendations (GR), which usually include item tokenizers and generative Large Language Models (LLMs), have demonstrated remarkable success across a wide range of scenarios. The majority of existing research efforts primarily concentrate on developing powerful item tokenizers or advancing LLM decoding strategies to attain superior performance. However, the critical fine-tuning step in GR frameworks, which is essential for adapting LLMs to recommendation data, remains largely unexplored. Current approaches predominantly rely on either the next-token prediction loss of supervised fine-tuning (SFT) or recommendationspecific direct preference optimization (DPO) strategies. Both methods ignore the exploration of possible positive unobserved samples, which is commonly referred to as the exposure bias problem. To mitigate this problem, this paper treats the GR as a multi-step generation task and constructs a GFlowNets-based fine-tuning framework (GFlowGR). The proposed framework integrates collaborative knowledge from traditional recommender systems to create an adaptive trajectory sampler and a comprehensive reward model. Leveraging the diverse generation property of GFlowNets, along with sampling and heuristic weighting techniques, GFlowGR emerges as a promising approach to mitigate the exposure bias problem. Extensive empirical results on two real-world datasets and with two different GR backbones highlight the effectiveness and robustness of GFlowGR.

cs.IR↗

arXiv2026

From Scaling to Structured Expressivity: Rethinking Transformers for CTR Prediction

Despite massive investments in scale, deep models for click-through rate (CTR) prediction often exhibit rapidly diminishing returns -- a stark contrast to the {predictable scaling laws} seen in large language models (LLMs). We identify the root cause as a {fundamental} \textit{structural misalignment}: {standard} Transformers assume sequential compositionality, whereas CTR data demand combinatorial reasoning over {heterogeneous} fields. To restore alignment, we introduce the \textbf{Field-Aware Transformer (FAT)}. {By reconstructing the standard Transformer block with field-centric parameters, FAT achieves \textit{structured expressivity}, {fundamentally shifting the model complexity dependence from the total vocabulary size $n$ with the number of fields $F$ ($n \gg F$).}} Crucially, to decouple model capacity from field cardinality, FAT employs a {Basis-Composed Hypernetwork} to synthesize field-specific parameters from shared bases, further reducing parameter complexity. {Theoretically, we ground this scaling behavior through a formal scaling law based on Rademacher complexity. Empirically, FAT outperforms exisiting state-of-the-art methods with up to \textbf{+4.38\%} AUC improvement, and delivers \textbf{+2.33\%} CTR and \textbf{+0.66\%} RPM in live production.} Our work establishes that scalable recommendation arises not from size alone, but from \textit{structured expressivity} -- architectural coherence with data semantics.

cs.IR↗

arXiv2026

Telecom quantum memory over one microsecond in nanophotonic lithium niobate

Nanophotonic quantum memory is a vital component for scalable quantum information processing for quantum computing, networking, and sensing applications. We store single-photon-level telecom-band optical pulses for more than a microsecond using an atomic frequency comb in erbium-doped thin-film lithium niobate, well beyond what is practically feasible via propagation in even the best nanophotonic devices due to propagation losses. We verify the quantum nature of this storage by demonstrating the phase coherence and sub-single-photon noise upon retrieval. We also show the flexibility of our platform by storing up to 20 temporal modes and demonstrating an acceptance bandwidth up to 2.2 GHz. These results establish erbium-doped thin-film lithium niobate as a practical platform for on-chip quantum memory at telecom wavelengths, a key missing element for photonic quantum computing and quantum networking.

quant-ph↗

arXiv2026

Co-Diffusion: An Affinity-Aware Two-Stage Latent Diffusion Framework for Generalizable Drug-Target Affinity Prediction

Predicting drug-target affinity is fundamental to virtual screening and lead optimization. However, existing deep models often suffer from representation collapse in stringent cold-start regimes, where the scarcity of labels and domain shifts prevent the learning of transferable pharmacophores and binding motifs. In this paper, we propose Co-Diffusion, a novel affinity-aware framework that redefines DTA prediction as a constrained latent denoising process to enhance generalization. Co-Diffusion employs a two-stage paradigm: Stage I establishes an affinity-steered latent manifold by aligning drug and target embeddings under an explicit supervised objective, ensuring that the latent space reflects the intrinsic binding landscape. Stage II introduces modality-specific latent diffusion as a stochastic perturb-and-denoise regularizer, forcing the model to recover consistent affinity semantics from noisy structural representations. This approach effectively mitigates the reconstruction-regression conflict common in generative DTA models. Theoretically, we show that Co-Diffusion maximizes a variational lower bound on the joint likelihood of drug structures, protein sequences, and binding strength. Extensive experiments across multiple benchmarks demonstrate that Co-Diffusion significantly outperforms state-of-the-art baselines, particularly yielding superior zero-shot generalization on unseen molecular scaffolds and novel protein families-paving a robust path for in silico drug prioritization in unexplored chemical spaces.

stat.ML↗

arXiv2026

Melting of quantum Hall Wigner and bubble crystals

A two-dimensional crystal melts via the proliferation and unbinding of topological defects, yet quantitatively predicting the melting temperature $T_m$ in real systems is challenging. Here we resolve this discrepancy in quantum Hall electron bubble phases by combining Corbino-geometry transport experiment in an ultraclean GaAs/AlGaAs quantum well for Landau levels 2 to 5 with Hartree--Fock elasticity and the full Kosterlitz--Thouless--Halperin--Nelson--Young melting criterion including the finite-temperature renormalization-group calculation. The theoretically obtained $T_m$ quantitatively captures the measured solid-liquid phase transition boundaries across all probed ranges, validating the bubble-crystal interpretation and establishing defect--mediated melting as a predictive framework for strongly interacting electronic solids. This agreement further supports using bulk transport to probe the energetics of topological defects and screening in quantum Hall physics, and the approach is readily extendable to other electronic crystals, including the generalized Wigner crystal in moiré Chern bands.

cond-mat.mes-hall↗

arXiv2026

MOON: Generative MLLM-based Multimodal Representation Learning for E-commerce Product Understanding

With the rapid advancement of e-commerce, exploring general representations rather than task-specific ones has attracted increasing research attention. For product understanding, although existing discriminative dual-flow architectures drive progress in this field, they inherently struggle to model the many-to-one alignment between multiple images and texts of products. Therefore, we argue that generative Multimodal Large Language Models (MLLMs) hold significant potential for improving product representation learning. Nevertheless, achieving this goal still remains non-trivial due to several key challenges: the lack of multimodal and aspect-aware modeling modules in typical LLMs; the common presence of background noise in product images; and the absence of a standard benchmark for evaluation. To address these issues, we propose the first generative MLLM-based model named MOON for product representation learning. Our method (1) employs a guided Mixture-of-Experts (MoE) module for targeted modeling of multimodal and aspect-specific product content; (2) effectively detects core semantic regions in product images to mitigate the distraction and interference caused by background noise; and (3) introduces the specialized negative sampling strategy to increase the difficulty and diversity of negative samples. In addition, we release a large-scale multimodal benchmark MBE for various product understanding tasks. Experimentally, our model demonstrates competitive zero-shot performance on both our benchmark and the public dataset, showcasing strong generalization across various downstream tasks, including cross-modal retrieval, product classification, and attribute prediction. Furthermore, the case study and visualization illustrate the effectiveness of MOON for product understanding. The data of our MBE benchmark is given in https://huggingface.co/datasets/Daoze/MM-Bench-E-Commerce.

cs.CV↗

arXiv2026

RAIR: A Rule-Aware Benchmark Uniting Challenging Long-Tail and Visual Salience Subset for E-commerce Relevance Assessment

Search relevance plays a central role in web e-commerce. While large language models (LLMs) have shown significant results on relevance task, existing benchmarks lack sufficient complexity for comprehensive model assessment, resulting in an absence of standardized relevance evaluation metrics across the industry. To address this limitation, we propose Rule-Aware benchmark with Image for Relevance assessment(RAIR), a Chinese dataset derived from real-world scenarios. RAIR established a standardized framework for relevance assessment and provides a set of universal rules, which forms the foundation for standardized evaluation. Additionally, RAIR analyzes essential capabilities required for current relevance models and introduces a comprehensive dataset consists of three subset: (1) a general subset with industry-balanced sampling to evaluate fundamental model competencies; (2) a long-tail hard subset focus on challenging cases to assess performance limits; (3) a visual salience subset for evaluating multimodal understanding capabilities. We conducted experiments on RAIR using 14 open and closed-source models. The results demonstrate that RAIR presents sufficient challenges even for GPT-5, which achieved the best performance. RAIR data are now available, serving as an industry benchmark for relevance assessment while providing new insights into general LLM and Visual Language Model(VLM) evaluation.

cs.IR↗

arXiv2026

NEZHA: A Zero-sacrifice and Hyperspeed Decoding Architecture for Generative Recommendations

Generative Recommendation (GR), powered by Large Language Models (LLMs), represents a promising new paradigm for industrial recommender systems. However, their practical application is severely hindered by high inference latency, which makes them infeasible for high-throughput, real-time services and limits their overall business impact. While Speculative Decoding (SD) has been proposed to accelerate the autoregressive generation process, existing implementations introduce new bottlenecks: they typically require separate draft models and model-based verifiers, requiring additional training and increasing the latency overhead. In this paper, we address these challenges with NEZHA, a novel architecture that achieves hyperspeed decoding for GR systems without sacrificing recommendation quality. Specifically, NEZHA integrates a nimble autoregressive draft head directly into the primary model, enabling efficient self-drafting. This design, combined with a specialized input prompt structure, preserves the integrity of sequence-to-sequence generation. Furthermore, to tackle the critical problem of hallucination, a major source of performance degradation, we introduce an efficient, model-free verifier based on a hash set. We demonstrate the effectiveness of NEZHA through extensive experiments on public datasets and have successfully deployed the system on Taobao since October 2025, driving the billion-level advertising revenue and serving hundreds of millions of daily active users.

cs.AI↗

arXiv2026

NOSA: Native and Offloadable Sparse Attention

Decoding throughput improvements from larger inference batches are limited by GPU memory, which is largely consumed by the key-value (KV) cache. Prior training-free KV cache offloading alleviates this by keeping redundant context on the CPU and fetching only a sparse subset for attention, but it often degrades long-generation quality due to training-inference mismatch on sparse patterns. Meanwhile, trainable sparse attention is incompatible with efficient offloading, as unconstrained KV accesses may force large CPU-to-GPU transfers and erase throughput gains. To this end, we propose NOSA, a trainable sparse attention mechanism natively designed for KV cache offloading. NOSA explicitly constrains the volume of CPU-GPU KV transfers, thereby achieving low communication overhead and high decoding throughput. We further build NOSI, a KV cache offloading inference system that fully unlocks NOSA's efficiency. Empirical results on 1,3,8B LLMs demonstrate that NOSA outperforms KV cache offloading baselines on general, long-input, and long-generation tasks, while boosting decoding throughput by up to 5.04x, 1.92x, and 1.83x over FullAttn, InfLLMv2, and ShadowKV, respectively. We release our code at https://github.com/thunlp/NOSA.

cs.CL↗

arXiv2026

LoFT-LLM: Low-Frequency Time-Series Forecasting with Large Language Models

Time-series forecasting in real-world applications such as finance and energy often faces challenges due to limited training data and complex, noisy temporal dynamics. Existing deep forecasting models typically supervise predictions using full-length temporal windows, which include substantial high-frequency noise and obscure long-term trends. Moreover, auxiliary variables containing rich domain-specific information are often underutilized, especially in few-shot settings. To address these challenges, we propose LoFT-LLM, a frequency-aware forecasting pipeline that integrates low-frequency learning with semantic calibration via a large language model (LLM). Firstly, a Patch Low-Frequency forecasting Module (PLFM) extracts stable low-frequency trends from localized spectral patches. Secondly, a residual learner then models high-frequency variations. Finally, a fine-tuned LLM refines the predictions by incorporating auxiliary context and domain knowledge through structured natural language prompts. Extensive experiments on financial and energy datasets demonstrate that LoFT-LLM significantly outperforms strong baselines under both full-data and few-shot regimes, delivering superior accuracy, robustness, and interpretability.

cs.LG↗