arXiv subjects

Jian Wu

Publications and source records attributed to Jian Wu.

At least 19 recordsLinked to original sources

arXiv2026

Counterexample-driven autonomous research with application to excess radiated energy in a Z-pinch

Recent breakthroughs in artificial-intelligence-assisted mathematics show that open-ended machine search can outperform human mathematicians, yet transferring this paradigm to physics requires refactoring the research workflow. We introduce Simjecture, an open-source framework in which language-model agents pursue human-defined physical conjectures. The framework tackles two central challenges in this transfer: first, because physics is empirical and no model is final, it structures inquiry around counterexample-driven falsification and prospective hypothesis repair; second, because physical theories must interact with the real world, it uses computational physics as an executable testbed while separating the qualification of numerical instruments from scientific claim review so solvers and diagnostics can be adapted within a recorded investigation. We apply the framework in an 82.6 h investigation to tackle the long-standing problem of radiated energy exceeding the imploding-shell kinetic energy in a Z-pinch. The investigation reveals competing effects of energy conversion, retention, and release timing. Stronger coherent helical deformation increases resistive conversion while reducing radiation and increasing retained energy. A broader trailing-plasma profile delays the later radiation maximum and recovers much of its early bolometric deficit, with a spectral response that changes during the pulse. These results provide new insight into energy conversion in high-energy-density plasma physics and demonstrate a new paradigm for physics research.

cs.AI↗

arXiv2026

GUI-AC: Enhancing Continual Learning in GUI Agents

Graphical User Interfaces (GUIs) serve as the dominant medium for human-computer interaction, yet building GUI agents that generalize across the vast diversity of real-world interface environments, with the same flexibility and robustness that humans naturally exhibit, remains unsolved. Notably, GUI data are inherently non-stationary: the continual emergence of previously unseen interface instances (e.g., novel domains and resolutions) induces persistent distribution shifts, significantly impeding the continual learning of existing GUI agents. Reinforcement fine-tuning (RFT) has attracted considerable attention as a promising approach. Nevertheless, RFT exhibits pronounced instability in its grounding capability, where rollout outcomes are highly imbalanced and fixed clipping can suppress valuable exploration. More importantly, policy probability alone does not reveal the semantic role of a grounding prediction. To address these challenges, we propose GUI-AC, a continual RFT framework that jointly characterizes grounding rollouts through advantage and policy probability. This yields a four-region view of grounding behavior and enables targeted optimization of Exploration and Overconfidence. GUI-AC further introduces grounding certainty to support two core mechanisms: (i) Adaptive Advantage, which strengthens the negative correction for overconfident grounding failures; and (ii) Dynamic Clipping, which relaxes the upper clipping bound for exploratory grounding rollouts. Extensive experiments show that GUI-AC consistently outperforms state-of-the-art baselines. Code is available at https://anonymous.4open.science/r/GUI-AC-Code.

cs.CV↗

arXiv2026

UserArena: Benchmarking Interactive User Simulation for Long-Horizon Shopping Agents

Agentic recommendation pursues shopper goals through sequences of context-dependent decisions. Developing and evaluating such policies requires faithful recommendation world models that capture both the interactive shopping environment and users' behavioral responses over long interactions. Existing resources rarely provide long-horizon, fine-grained real-user trajectories or support trajectory-level evaluation and offline validation against production A/B results. We introduce UserArena, a real-world e-commerce interactive benchmark containing 64,729 session trajectories from 11,010 users, spanning 566,102 items and 13 action types. UserArena pairs these trajectories with an interactive environment and supports offline validation against outcomes from a controlled production A/B test, enabling five-level evaluation from output validity and local decisions to sequential behavior, personalization, and outcomes. To improve simulator fidelity over these long-horizon interactions, we further propose On-Policy Trajectory Distillation (OPTD), which uses trajectory-level intent to provide dense teacher targets on student-generated rollouts. Experiments show that OPTD achieves the strongest overall fidelity among the evaluated simulators under UserArena's evaluation protocol and most closely reproduces the production policy effect in offline simulation.

cs.IR↗

arXiv2026

Optimal weighted sampling of positive and normal operator orbits

We study weighted sampling of operator orbit frames in infinite dimensional separable Hilbert spaces $H$, keeping the operator and generating vectors fixed. Let $N_μ(R)$ count the sampling times in $[0,R]$. For a single orbit of a positive operator, let $m_A(R)$ count the eigenvalues in $(0,e^{-1/R}]$. Every real sampling design that preserves the frame property must satisfy the lower bound $N_μ(R)\ge m_A(R)-O_μ(1)$. Conversely, for every $η>0$, we construct an integer sampling design that preserves the frame property and satisfies $N_μ(R)\le(1+η)m_A(R)+O_η(1)$. Thus the coefficient $1$ is optimal. Under additional spectral conditions, we attain $N_μ(R)=m_A(R)+O(1)$ and characterize the weights that preserve the frame property on the constructed sets. For finitely or countably many orbits of a positive operator, let $P_j$, $j\ge1$, be the spectral projection corresponding to $2^{-j-1}<-\logλ\le2^{-j}$ and define $K_A(R)=\#\{j\ge1:2^j\le R,\ P_j\ne0\}$. Every real sampling design that preserves the frame property satisfies $N_μ(R)\ge c_μK_A(R)-C_μ$ for $R\ge2$, with constants $c_μ>0$ and $C_μ\ge0$. Conversely, we construct an integer sampling design that preserves the frame property and satisfies $N_μ(R)\le L K_A(R)+L$ for $R\ge2$, where $L$ depends only on the bounds of the given orbit frame. For normal operators of norm one, we prove that the smallest frame condition number is attained among integer sampling designs that preserve the frame property and satisfy a prescribed counting bound, whenever this class is nonempty. Finally, we construct one normal operator with two families of generating vectors for which the optimal growth orders of $N_μ(R)$ are $\log R$ and $R$, respectively, for both real and integer sampling.

math.FA↗

arXiv2026

Sep-Latent: Separator Tokens as Natural Compression Anchors for Implicit Reasoning

Large Language Models (LLMs) have demonstrated remarkable success in complex reasoning through explicit Chain-of-Thought (CoT), but the associated computational overhead and token verbosity remain significant barriers to efficiency. While implicit reasoning in continuous latent spaces offers a promising alternative, existing methods like Coconut and CODI often rely on fixed-length latent sequences and suffer from representational collapse. In this paper, we propose Sep-Latent, an adaptive implicit reasoning framework that utilizes separator tokens as structural semantic anchors. By introducing an Attention-Decay mechanism, we force the model to aggregate and compress reasoning logic into these anchors during training. We then align explicit and implicit pathways through multi-level distillation across all layers to ensure logical consistency. Crucially, Sep-Latent enables dynamic reasoning depth by adaptively terminating the latent thinking phase based on query complexity. Experimental results on the GSM8k-Aug benchmark show that Sep-Latent achieves an accuracy of 56.8\% with an average of only 5.4 tokens, significantly outperforming prior fixed-length implicit models in efficiency while narrowing the performance gap with explicit CoT. Further analysis via PCA and symbolic variant testing confirms that our method ensures semantic grounding and exhibits superior robustness to perturbations.

cs.AI↗

arXiv2026

Universal weighted sampling of positive and normal operator orbits

We study common weighted sampling designs for operator systems that are exactly observable at nonnegative integer times. For positive operators with original frame condition number at most $κ$, we construct a common integer design with arbitrarily small relative Gramian error and $O(\log R)$ distinct times up to $R$. The optimal logarithmic counting coefficient for preserving the frame property satisfies $C^*(κ)\simπ^{-2}\logκ$ as $κ\to\infty$. Real and integer times give the same infimum, and we determine its asymptotics as $κ\downarrow1$. A finite bound on the sampled condition number strictly increases this coefficient. For tight frames, we determine the exact optimum under this constraint and construct an integer design attaining it. Without this constraint, the infimum is never attained. For positive operators, a design valid for every finite $κ$ requires counting faster than $\log R$, with arbitrarily slow divergence of the ratio. For unrestricted normal operators and fixed $κ>1$, at least linear counting is necessary. Under the logarithmic sector condition $|\argλ|\le M(-\log|λ|)$ with fixed $M\ge0$, logarithmic sampling is possible with a sharp leading coefficient. For fixed Borel spectral sets satisfying a radial density condition, we characterize logarithmic sampling and determine the minimum upper counting exponent, attained by an integer design. Counterexamples show that the radial condition cannot be omitted.

math.FA↗

arXiv2026

Density and Spectral Criteria for Universal Nonuniform Sampling of Positive Operator Orbits

We characterize locally finite sets $T\subset[0,\infty)$ that preserve exact observability from nonnegative integer times for every bounded positive operator $A$ and bounded observation operator $C$. The Stein identity reduces universality to uniform upper and lower bounds on radial Stein--Pick cones over all finite spectral models. Writing $N_T(R)=\#(T\cap[0,R])$, we prove that universal Bessel preservation is equivalent to $N_T(R)=O(R)$. The conditions $0\in T$ and $N_T(R)\asymp R$ characterize universal observability preservation under $[A,C^*C]=0$, as well as uniform bound preservation for each fixed number of spectral values. However, we construct a clustered set satisfying these conditions whose optimal sampled lower bounds tend to zero as the spectral count grows, despite fixed integer observability bounds. The same set destroys the frame property of a singly generated positive diagonal Carleson frame while preserving its Bessel property. When natural density exists, universality is equivalent to $0\in T$ and $0<d(T)<\infty$. More generally, containing zero and a subset of positive finite natural density suffices under linear upper counting. For sets containing zero, universality is invariant under sublinear differences of counting functions. For general normal operators, universal Bessel preservation is instead equivalent to uniformly bounded counts on unit intervals; under a fixed logarithmic sector condition, it is again characterized by linear upper counting.

math.FA↗

arXiv2026

Experimental certification of multipartite Bell correlations using only few-body symmetric correlations

Certifying Bell correlations in increasingly larger multipartite quantum systems is challenging because conventional Bell inequalities rely on many-body correlations, which become experimentally demanding to estimate as system size grows. Here we demonstrate that multipartite Bell correlations can be certified using few-body permutation-invariant (PI) correlations with a fixed maximal correlation order. We experimentally test few-body PI Bell correlation witnesses tailored for six- and eight-photon postselected Dicke states. The addition of a few four-body correlators is found to significantly enhance the noise tolerance compared with previously studied two-body witnesses: for instance, for six qubits, the selected witness tolerates a white-noise fraction of $p_c\approx0.301$, compared with $p_c\approx0.048$ for the best two-body PI inequality with the same two settings per party. For six and eight photons respectively, we observe violations of the witness bounds by $32$ and $8.9$ standard deviations, arising from postselected polarisation states with fidelities of $0.933(4)$ and $0.916(11)$ to the corresponding half-excitation Dicke states. The witnesses for the six- and eight-photon cases use the same symmetric two- and four-body correlators, just with different coefficients. These results demonstrate an experimentally accessible approach for certifying multipartite Bell correlations in larger systems without requiring full-body correlation measurements.

quant-ph↗

arXiv2026

ReLMem: Learning Recurrent Memory for Longitudinal EHR Modeling

Longitudinal electronic health record (EHR) modeling requires integrating new visits with an expanding patient history. Yet the continual accumulation of clinical information imposes increasing computational and memory costs on large language models (LLMs) when they process and retain complete patient histories. A practical alternative is visit-wise recurrent compression, which incorporates each incoming visit into a compact, continually updated patient memory. However, under a fixed memory budget, successive updates must integrate new information without progressively losing critical historical evidence needed to subsequent tasks. To address this challenge, we introduce Recurrent Longitudinal Memory (ReLMem), a framework that learns to maintain fixed-capacity patient memory for efficient downstream prediction with a frozen LLM. ReLMem equips this LLM with lightweight compression adapters to recurrently update the memory from its previous state and each incoming visit, without rereading earlier records. Specifically, we develop a multi-granularity optimization strategy to preserve task-relevant information throughout recurrent updates and support downstream prediction from the final memory. The intermediate supervision aligns attention outputs from compressed memory and the full history under identical queries, while prediction supervision minimizes cross-entropy with ground truth answers conditioned on the final memory. On EHR-based medication prediction, ReLMem approaches the F1 scores of full-history baseline while reducing average retained historical storage by 97.1%. Under the same memory budget, it improves macro- and micro-F1 over the strongest compressed-memory baseline by 4.66 and 4.75 percentage points, respectively. These results highlight the value of learning recurrent patient memory for efficient longitudinal EHR modeling.

cs.LG↗

arXiv2026

VoxelSage: Tool-Augmented 3D CT Analysis and Simulator-Shielded Sequential Resection Planning for Liver Tumors

Preoperative liver-tumor assessment requires segmentation, physical-space measurement, visual evidence, and resection planning from the same three-dimensional CT volume. Existing tools often handle these steps separately, while language models cannot reliably compute physical measurements from CT. To provide an integrated workflow, we present VoxelSage, a multi-modal system for two- and three-dimensional visualization, liver-tumor analysis, and preoperative resection planning. Its dual-port architecture separates language-model orchestration from image computation: Port A interprets requests and selects skills, while Port B applies them to CT volumes and segmentation masks and returns structured results. Keeping physical measurements in Port B prevents the LLM from computing them directly and reduces the risk of fabricated numerical results. Eight built-in skills support quantitative analysis, visual evidence generation, three-dimensional reconstruction, segmentation refinement, and sequential resection planning; user-defined skills can extend these functions. For sequence planning, a behavior-cloned neural ranker orders candidate resection targets, while a simulator-based shield checks them against predefined constraints. Across 256 unseen simulator scenes, this approach reduced mean simulated time from 34.274 to 33.388 min (0.886 min, 2.59%) and mean simulated blood loss from 300.847 to 183.852 mL (116.995 mL, 38.89%) relative to a deterministic baseline. These results demonstrate system integration and simulator-level performance, not clinical efficacy or safety. The public implementation is available at https://github.com/ZJUMAI/VoxelSage.

cs.CV↗

arXiv2026

Loss-Resilient Quantum Networking with Dicke Entanglement

Multipartite entanglement can endow a quantum network with capabilities unavailable classically, yet their evolution under progressively reduced physical and informational access remains largely unexplored. Here we realize the first eight-photon half-excitation Dicke state, with a fidelity of 0.922(11), and use it to implement two distinct network tasks: open-destination teleportation and quantum telecloning. An unknown state can be teleported from one agent to any of seven remote agents with fidelities of 0.926(7)-0.941(6), while seven approximate clones are distributed with a fidelity of 0.713(3), close to the optimal value of 5/7. The protocol remains above the classical limit under imperfect photon detection, transmission loss and missing information from participating agents, and outperforms the corresponding GHZ benchmarks. As fewer agents contribute, the teleportation fidelity falls through a stepwise hierarchy whose two extremes recover open-destination teleportation and optimal telecloning, connecting both tasks within a single operational framework. These results establish Dicke entanglement as a promising architecture for loss-resilient quantum networks, opening a route towards quantum technologies that retain useful functionality under realistic operating conditions.

quant-ph↗

arXiv2026

Characterizations and boundary regularity of continuous preservers of Carleson interpolating sequences

We characterize the continuous mappings of the unit disk that preserve every Carleson interpolating sequence. Preservation in one direction is equivalent to being a disk homeomorphism whose inverse is uniformly continuous in the pseudohyperbolic metric and whose associated weighted pushforward is bounded on positive Carleson measures. Preservation in both directions is equivalent to being a disk homeomorphism that, together with its inverse, is uniformly continuous in that metric and has a strongly quasisymmetric boundary extension. Within this class, comparability of the boundary defects is equivalent to a bi-Lipschitz boundary map. Building on the canonical factorization established in our earlier work, we give explicit criteria for membership in the continuous boundary kernel. For a homeomorphism of the closed disk with identity boundary values and a uniform boundary expansion of first order, kernel membership is equivalent to positivity of the determinant of the real boundary differential, or equivalently of the inward normal derivative of the boundary defect. No differentiability in the interior is required. The $C^1$ criterion on the closed disk follows as a corollary, while a counterexample shows that pointwise boundary differentiability does not suffice. We also characterize factorizations with a quasiconformal kernel and record the exact maximal dilatation of its radial factor.

math.CV↗

arXiv2026

Carleson Interpolation, Boundary Rigidity, and Classification of Spectral Transformations Preserving Diagonal Operator Orbit Frames

Maps $Φ:D\to D$ of the unit disk preserve frame generator sets for single orbits of normal diagonal operators exactly when they preserve Carleson interpolating sequences bidirectionally and satisfy $1-|Φ(z)|^2\asymp1-|z|^2$ uniformly. Denote this class by $\mathcal P$. Equality of frame generator sets forces the boundary estimate, since every frame generator $f=(f_j)$ satisfies $|f_j|^2\asymp 1-|λ_j|^2$. No continuity, measurability, or analyticity is assumed; Carleson preservation alone remains flexible and, to our knowledge, unclassified in general. Equivalently, normalized Szego kernel Riesz sequences are preserved bidirectionally, with uniformly comparable unnormalized kernel norms at $z$ and $Φ(z)$. Equivalently, we require uniform two-sided Carleson norm bounds for ordinary pushforwards of finite positive discrete measures and bidirectional preservation of pairwise pseudohyperbolic separation. Every $Φ\in\mathcal P$ has a canonical bilipschitz radial trace $h_Φ(ζ)=\lim_{r\to1^-}Φ(rζ)$ on $T=\partial D$, with uniform convergence. We give an intrinsic geometric characterization of $\mathcal K=\{Ψ\in\mathcal P:h_Ψ=\mathrm{id}\}$ and prove the unique factorization $Φ=Ψ\circ E_{h_Φ}$, $Ψ\in\mathcal K$, where $E_h(rζ)=rh(ζ)$ and $E_h(0)=0$; thus $\mathcal P\cong\mathcal K\rtimes\mathrm{BiLip}(T)$ as monoids. For $m$ orbits, $\mathcal P_m=\mathcal P$ for every finite $m\ge1$, whereas $\mathrm{Aut}(D)\subsetneq\mathcal P_ω\subsetneq\mathcal P$. Countable-orbit preservers in $\mathcal P_ω$ are uniform pseudohyperbolic homeomorphisms. If $φ\in\mathrm{Aut}(D)$ and $|Φ(z)-φ(z)|=o(1-|z|)$ uniformly as $|z|\to1$, then $Φ\in\mathcal P_ω$ exactly when $Φ$ is a disk homeomorphism. Two-sided tests characterize $\mathcal P_ω$. Holomorphic members of these classes are exactly disk automorphisms.

math.FA↗

arXiv2026

Knowledge-Enriched Structured EHR Features for 30-Day Hospital Readmission Prediction on MIMIC-IV

Recent approaches to 30-day hospital readmission prediction rely on pre-trained language models applied to discharge summaries. Although these methods achieve strong performance, they depend on the availability of clinical notes, incur substantial computational costs, and yield representations that lack interpretability. We propose a knowledge-enriched feature representation that augments structured Electronic Health Record (EHR) data with four medical knowledge sources: disease ontology mapping, procedure classification, drug ingredient vocabulary, and organ system laboratory aggregation, without using clinical notes. Each feature dimension corresponds to a named clinical concept, yielding a sparse and interpretable patient representation. The approach is evaluated with six classifiers on a MIMIC-IV v2.2 cohort. Under 20-fold cross-validation, the best configuration achieves an AUROC of 0.743. This performance is comparable to that of previously reported methods on this dataset, including both those using only structured data and those incorporating clinical notes, while requiring considerably less computational cost. Interpretability analysis shows that demographics, organ system labs, drug ingredient features, and first-level ontology disease categories drive prediction, while deeper hierarchy levels contribute negligibly. These findings indicate that knowledge-enriched structured features offer a competitive and efficient alternative to embeddings from clinical notes for 30-day readmission prediction.

cs.LG↗

arXiv2026

In-Hospital Stroke Risk-State Classification from PPG-Derived Hemodynamic Features

The scarcity of temporally aligned pre-event physiological data limits the study of stroke risk states before documented clinical recognition. We focus on patients who experienced stroke during hospitalization while undergoing continuous monitoring, enabling retrospective analysis of pre-anchor photoplethysmography (PPG). Using MIMIC-III and MC-MED, an LLM-assisted pipeline generated candidate stroke anchors from unstructured notes. All retained anchors underwent physician adjudication, and a stratified 100-case double-review audit showed 95.0% candidate-to-adjudicated agreement within +/-15 min. We identified 176 MIMIC-III patients and 154 MC-MED patients with eligible synchronized pre-anchor PPG. A fixed 17-channel hemodynamic representation and ResNet-1D classifier were evaluated using patient-level five-fold internal validation and frozen external testing. At validation-selected operating points, F1-scores were 0.7956, 0.8759, and 0.9406 for 4-, 5-, and 6-hour horizons in MIMIC-III and 0.9256, 0.9595, and 0.9888 in MC-MED; the corresponding threshold-free AUCs ranged from 0.6124 to 0.7079. The PPG model achieved higher F1 than four non-waveform clinical and structured-EHR comparators in all six cohort-horizon settings. On high-risk non-stroke controls, window-level false-positive rates ranged from 0.1371 to 0.2938 and decreased to 0.0183-0.1739 after persistence aggregation. These retrospective findings support measurable pre-anchor PPG structure, but do not establish biological stroke onset, a calibrated bedside alarm, or a clinically validated prediction lead time.

cs.LG↗

arXiv2026

LoopMemGR: From Behavior Logs to Evolving Memory for Generative Recommendation

Generative recommendation formulates next-item prediction as conditional autoregressive generation over discrete Semantic IDs, enabling end-to-end recommendation over large-scale item spaces. However, most existing methods follow a history-as-context paradigm that repeatedly reconstructs user preference from behavior history while discarding system-side recommendation decisions after each request. This creates an asymmetric memory: the system remembers what the user has done, but not what it has previously recommended or learned from the resulting feedback. Consequently, useful preference-validation signals, potential negative evidence, and historical exploration information cannot be directly reused across requests. To address these limitations, we propose LoopMemGR, a closed-loop recommendation experience memory framework for generative recommendation. In addition to the conventional behavior log, LoopMemGR maintains a recommendation experience log that records past recommendation--feedback trajectories. It extracts request-relevant evidence through three complementary views: the recency view captures short-term interaction dynamics, the frequency view summarizes recurring recommendation patterns, and the global view distills transferable regularities shared across users. These signals are compressed into a fixed number of experience tokens to condition the generative backbone under a bounded input budget. Extensive experiments on an industrial Taobao dataset demonstrate the effectiveness of closed-loop experience accumulation and multi-view experience extraction.

cs.IR↗

arXiv2026

URL Extraction from Scholarly Documents: A Cross-Format Comparative Analysis

URLs in scholarly documents link to rich external resources such as datasets, software, publications, and websites. Extracting these URLs is crucial in the data preparation stage of many downstream tasks, such as link rot analysis, web crawling, and building knowledge graphs. However, existing studies often downplay this phase, simply extracting URLs from a single format, usually text directly converted from PDFs. We present a systematic study evaluating URL extraction across six input formats (text with annotation layer, LaTeX, HTML, XML, Markdown, and PNG converted from PDF). To support the evaluation, we compiled a benchmark dataset consisting of 2,338 manually annotated URLs from 200 arXiv papers spanning a wide range of domains over a 33-year period. In addition to evaluating individual file formats, we also compared 63 composite input-format combinations. Our extensive evaluations indicate that TEXTWAL achieves the best performance among single-format inputs, while TEXTWAL+LaTeX achieves the best overall URL extraction performance. The same trend is observed for URLs linking to open-access datasets and software. To further validate these findings, we apply our format-specific URL extraction pipelines to a longitudinal random sample of 364,744 arXiv papers spanning 33 years. We observe a sharp increase in URL density after 2015, along with remarkable differences in URL extraction across file formats over time. Overall, our study highlights the importance of selecting an appropriate format for URL extraction from scholarly documents. The dataset and code are publicly available at: https://github.com/lamps-lab/arxiv-url-bench .

cs.DL↗

arXiv2026

SPARC: Sequence-aware Progressive Attribute Routing and Compression Framework for Generative Recommendation

Generative recommendation tokenizes items as discrete Semantic IDs (SIDs) and autoregressively generates target items from users' historical SID sequences. Although existing SIDs incorporate multimodal and structured information, they are typically statically assigned and independent of the current interaction context. In industrial scenarios, each behavior also contains heterogeneous attributes, such as category, brand, price, behavior type, and timestamp. Fully expanding these features greatly increases the input length, while directly compressing them into a single representation may prematurely discard context-relevant information. We propose \textbf{SPARC}, \uline{\textbf{S}}equence-aware \uline{\textbf{P}}rogressive \uline{\textbf{A}}ttribute \uline{\textbf{R}}outing and \uline{\textbf{C}}ompression Framework for Generative recommendation. SPARC first models the sequential dependencies of each field type to obtain context-aware field representations. It then routes the original, contextual, and identity representations of different fields into multiple slots to preserve complementary information under a fixed capacity. Finally, lightweight cross-item interaction integrates the intermediate tokens and compresses each historical item into a single token. Following the principle of contextualizing before compression, SPARC enriches user-history representations without increasing the input length of the generative backbone. Experiments on industrial Taobao and public Amazon datasets demonstrate that SPARC outperforms strong conventional and generative baselines. Further comparisons with static compression variants show that the improvement of SPARC comes from context-conditioned information retention rather than merely increasing the expressiveness of the compression module.

cs.IR↗