Searcharxiv⌕ Search

arXiv subjects

Chengfu Huo

Publications and source records attributed to Chengfu Huo.

17 recordsLinked to original sources
arXiv2026

Learning from What You Retrieve: Online RL Fine-Tuning for Semantic Retrieval

In large-scale e-commerce retrieval, dual-encoder retrievers are op- timized for contrastive similarity, whereas downstream rerankers capture finer-grained relevance preferences; this objective mis- match limits end-to-end retrieval quality. Reinforcement Learning offers a way to use reward-model feedback for retriever adaptation, but we observe that standard policy-gradient updates can degrade embedding geometry, especially when the document index must remain frozen due to industrial constraints. To address this, we propose PAO (Positive-Advantage-Only), a selective RL optimization method. Our analysis reveals that in- discriminate penalization of negative samples (pushing away) in a frozen high-dimensional space disrupts pre-trained semantic man- ifolds. PAO selectively applies gradient updates only to retrieved items with positive advantages, effectively pulling query embed- dings toward high-reward regions while preserving global topo- logical stability. Experiments on both a massive industrial dataset and public benchmarks demonstrate that PAO significantly outper- forms standard RL and distillation baselines.

cs.IR↗
arXiv2026

CommerceVibe: Learning to Design E-Commerce Creatives as Executable Visual Code via Dual-Feedback Reinforcement Learning

High-quality e-commerce creatives are essential for presenting products and conveying marketing messages. Recent diffusion models enable scalable creative generation and produce visually compelling images, but their flattened raster outputs often contain distorted text and inconsistent product details, requiring refinement before deployment. Moreover, without explicit structure, the resulting creatives are difficult to edit and reuse, while complex design requirements remain challenging to encode as verifiable training signals. To address these challenges, we present CommerceVibe, which represents creatives as executable visual code and formulates generation as conditional HTML/CSS program synthesis. Given product images, design requirements, and product information, it produces renderable, editable, and reusable creatives. We further introduce dual-feedback reinforcement learning, in which rule-based feedback evaluates rendered programs for text readability, product visibility, and layout validity, while visual feedback from a vision-language model (VLM) assesses rendered creatives against input specifications across six perceptual and commercial dimensions. Together, these complementary feedback signals improve both constraint satisfaction and perception-dependent quality. We perform supervised fine-tuning (SFT) of Qwen3.5-9B on over 28,000 e-commerce examples, followed by dual-feedback reinforcement learning. On a 1,300-case benchmark, the optimized CommerceVibe model achieves a weighted score of 94.0/100, compared with 87.3 for the SFT-only variant, and outperforms strong external models. Blind evaluations by five e-commerce design experts further validate these improvements. CommerceVibe supports controllable, editable, and scalable e-commerce creative production.

cs.CV↗
arXiv2026

Bidirectional Context Self-Distillation for Reinforcement Learning of Skill-Based LLM Agents

External natural-language skills provide large language model (LLM) agents with reusable and editable guidance for solving complex tasks. Yet their effectiveness depends not only on skill quality, but also on whether the policy can translate the provided guidance into appropriate actions. However, methods specifically designed to improve this skill-utilization ability remain largely underexplored. In practice, skill-based agents are commonly trained with reinforcement learning objectives centered on task-level rewards, which offer limited supervision and struggle to capture subtle differences in how effectively the policy uses the provided skills. We propose BCSD (Bidirectional Context Self-Distillation), a framework that combines self-distillation with reinforcement learning to train LLM agents to use external skills more effectively. Unlike prior self-distillation methods that rely on a single privileged context, BCSD evaluates each trajectory from two complementary skill-context views. The augmented view introduces higher-level Meta-Skill guidance, while the reduced view prunes general guidance to highlight task-specific skills. Their complementary token-level signals are combined to rescale the RL advantage. Experiments on ALFWorld and WebShop demonstrate that BCSD achieves the strongest overall performance across model scales, enabling agents to utilize external skills more effectively. Ablation studies further verify the complementary contributions of the augmented and reduced context views. Code will be released to ensure full reproducibility.

cs.AI↗
arXiv2026

OmniPack: Unified Token Compression for Efficient Omni-modal Large Language Models

Omni-modal large language models (Omni-LLMs) have achieved remarkable performance on audio-visual understanding tasks, but processing long and highly redundant visual and audio token sequences incurs substantial computational overhead, demanding aggressive token compression for efficient deployment. Existing methods often degrade at low token budgets: pre-LLM compression may discard structurally important and globally distributed evidence, whereas inner-LLM compression often underexploits query-conditioned audio-visual collaboration. To address these limitations, we propose OmniPack, a training-free framework that coordinates structural compression before the LLM with task-relevant semantic refinement within the LLM. Before the LLM, OmniPack removes structural redundancy through modality-specific importance, global coverage, and similarity-aware merging. After sufficient multimodal interaction, it further consolidates diverse, task-relevant representations through textual guidance and audio-visual collaboration. Extensive experiments on five benchmarks with three Omni-LLM backbones demonstrate that OmniPack consistently achieves the best performance-efficiency trade-off across diverse retention ratios, outperforming all existing methods. Notably, on Qwen2.5-Omni-7B, OmniPack preserves 98.0% of the original performance while reducing FLOPs to 16.7%, and still retains 92.9% of the original performance with only 6.8% of the original FLOPs.

cs.CV↗
arXiv2026

Beyond Solution-Centric Search: Adaptive Inquiry and Knowledge Revision for Autonomous ML Engineering

Long-horizon autonomous research tasks such as machine learning engineering require systems to make interdependent decisions under a limited budget. Existing LLM-based agents typically organize candidate-solution improvement through tree, graph, or chain structures, meaning that the search process determines how information is acquired and managed. We call this design solution-centric search and propose instead the information paradigm, in which an evolving information state represents the system's understanding of the task and guides solution improvement. We instantiate this paradigm in Iris, an inquiry-revision loop. For information acquisition, Iris generates local action plans from the current information state and uses epistemic actions to probe decision-critical unknowns without modifying the retained solution. For information management, Iris synthesizes observations across experiments into task knowledge composed of revisable claims with explicit scope and status. It updates this knowledge as new evidence arrives and constructs each decision context from raw evidence, structured summaries, or task knowledge at the required level of detail. On MLE-Bench, Iris attains a 64.9% any-medal rate under a 12-hour budget, the highest among compared systems. Across four tasks spanning harness engineering and model post-training, Iris also demonstrates cross-domain generalization.

cs.AI↗
arXiv2026

MerchantBench: Benchmarking LLM Agents for Long-Term Coherence in E-Commerce Operations

Large language model agents are increasingly evaluated as autonomous tool users, yet most benchmarks focus on bounded tasks with immediate success criteria. Real-world deployments often require Long-Term Coherence, the capacity to preserve purposeful behavior across extended horizons while adapting decisions to accumulated evidence. Evaluating this capacity requires a persistent environment in which actions constrain future choices, feedback arrives at heterogeneous delays, and incoherent behavior produces measurable cumulative effects. Seller-side e-commerce provides a suitable setting for this evaluation through recurrent and interdependent decisions over Product Sourcing, Listing and Pricing Control, Cash-Flow Management, and Mixed-Latency Feedback Adaptation. We introduce MerchantBench, a 365-day order-level simulation grounded in 98,843 real e-commerce product records and equipped with 26 tools for agent interaction. MerchantBench couples promptly observable Upstream Supplier Events with delayed Downstream Order Outcomes, requiring agents to follow individual order lifecycles and revisit earlier decisions. We evaluate eight LLMs under two agent frameworks in 48 runs, each spanning 365 simulated days. Our results reveal a substantial gap between even the latest LLMs and human participants, with the best LLM configuration attaining only 27.3\% of the mean final net assets achieved by human participants.

cs.AI↗
arXiv2026

IndustryBench-MIPU: Benchmarking Multi-Image Attribute Value Extraction for Industrial Products

Industrial products such as valves and circuit breakers are defined by dense technical specifications that govern procurement, compatibility, and safety across supply chains. These specifications are scattered across multiple heterogeneous product images, including specification tables, nameplates, and technical drawings, yet whether Multimodal Large Language Models (MLLMs) can reliably recover them remains underexplored. To fill this gap, we introduce IndustryBench-MIPU, the first large-scale benchmark for multi-image industrial product understanding, built around structured attribute extraction -- recovering property-value pairs from product images. This task jointly probes text recognition on specification tables and nameplates, visual reasoning over technical drawings, domain knowledge to decode industrial terminology, and cross-image evidence integration to assemble scattered specifications. Concretely, the benchmark comprises 4,559 products across 27,652 images with 103,703 annotations spanning 18 industrial categories, constructed through multi-model consensus and three-tier quality assurance. Evaluating nine MLLMs under both single-image and product-level multi-image settings reveals a stark completeness gap: models achieve high precision (86--94%) but the best recovers only 49.9% of product-level attributes; moving from single-image to multi-image extraction costs 15--34 percentage points of recall. Multi-image completeness, not single-image accuracy, is the core bottleneck. Dataset and code are publicly available.

cs.CV↗
arXiv2026

ARBOR: Online Process Rewards via a Reusable Rubric Buffer for Search Agents

LLM-based search agents are trained predominantly with outcome-only reward, leaving the search process itself unsupervised. This signal degenerates on outcome-homogeneous groups where all sampled trajectories share the same correctness, yielding zero within-group advantage and no gradient. Existing process supervision either trains a costly verifier or generates per-query rubrics that are inconsistent across queries and discarded after one use. We propose ARBOR (Adaptive Rubric Buffer for Online Reward), a reusable process-reward framework that maintains a rubric memory shared across queries. Query-local drafts induced from contrastive trajectories are admitted, consolidated into cross-query common rubrics, and retired as the policy evolves. A small active subset of common rubrics scores trajectories via sparse pairwise judging, and the resulting scores are added to the base reward, providing process-level gradient even when outcome reward is uniform. ARBOR consistently outperforms GRPO and DAPO baselines on four multi-hop QA benchmarks, raising average LLM-judge accuracy by up to 4.2 points and converting up to 42% of otherwise-zero-gradient training groups into informative ones.

cs.CL↗
arXiv2026

IndustryBench: Probing the Industrial Knowledge Boundaries of LLMs

In industrial procurement, an LLM answer is useful only if it survives a standards check: recommended material must match operating condition, every parameter must respect a regulated threshold, and no procedure may contradict a safety clause. Partial correctness can mask safety-critical contradictions that aggregate LLM benchmarks rarely capture. We introduce IndustryBench, a 2,049-item benchmark for industrial procurement QA in Chinese, grounded in Chinese national standards (GB/T) and structured industrial product records, organized by seven capability dimensions, ten industry categories, and panel-derived difficulty tiers, with item-aligned English, Russian, and Vietnamese renderings. Our construction pipeline rejects 70.3% of LLM-generated candidates at a search-based external-verification stage, calibrating how unreliable industrial QA remains after LLM-only filtering. Our evaluation decouples raw correctness, scored by a Qwen3-Max judge validated at $\kappa_w = 0.798$ against a domain expert, from a separate safety-violation (SV) check against source texts. Across 17 models in Chinese and an 8-model intersection over four languages, we find: (i) the best system reaches only 2.083 on the 0--3 rubric, leaving substantial headroom; (ii) Standards & Terminology is the most persistent capability weakness and survives item-aligned translation; (iii) extended reasoning lowers safety-adjusted scores for 12 of 13 models, primarily by introducing unsupported safety-critical details into longer final answers; and (iv) safety-violation rates reshuffle the leaderboard -- GPT-5.4 climbs from rank 6 to rank 3 after SV adjustment, while Kimi-k2.5-1T-A32B drops seven positions. Industrial LLM evaluation therefore requires source-grounded, safety-aware diagnosis rather than aggregate accuracy. We release IndustryBench with all prompts, scoring scripts, and dataset documentation.

cs.AI↗
arXiv2026

How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning

Thinking LLMs produce reasoning traces before answering. Prior activation steering work mainly targets on shaping these traces. It remains less understood how answer tokens actually read and integrate the reasoning to produce reliable outcomes. Focusing on quantitative reasoning, we analyze the answer-to-reasoning attention and observe a benign self-reading pattern aligned with correctness, characterized by a forward drift of the reading focus along the reasoning trace and a persistent concentration on key semantic anchors, whereas incorrect solutions exhibit diffuse and irregular attention pattern. We interpret this as internal certainty during answer decoding, where the model commits to a viable solution branch and integrates key evidence. Following this, we propose a training-free steering method driven by Self-Reading Quality (SRQ) scores combining geometric metrics for process control with semantic metrics for content monitoring. SRQ selects data to build steering vectors that guide inference toward benign self-reading and away from uncertain and disorganized reading. Experiments show that our method yields consistent accuracy gains.

cs.CL↗
arXiv2024

Knowledge Enhanced Multi-intent Transformer Network for Recommendation

Incorporating Knowledge Graphs into Recommendation has attracted growing attention in industry, due to the great potential of KG in providing abundant supplementary information and interpretability for the underlying models. However, simply integrating KG into recommendation usually brings in negative feedback in industry, due to the ignorance of the following two factors: i) users' multiple intents, which involve diverse nodes in KG. For example, in e-commerce scenarios, users may exhibit preferences for specific styles, brands, or colors. ii) knowledge noise, which is a prevalent issue in Knowledge Enhanced Recommendation (KGR) and even more severe in industry scenarios. The irrelevant knowledge properties of items may result in inferior model performance compared to approaches that do not incorporate knowledge. To tackle these challenges, we propose a novel approach named Knowledge Enhanced Multi-intent Transformer Network for Recommendation (KGTN), comprising two primary modules: Global Intents Modeling with Graph Transformer, and Knowledge Contrastive Denoising under Intents. Specifically, Global Intents with Graph Transformer focuses on capturing learnable user intents, by incorporating global signals from user-item-relation-entity interactions with a graph transformer, meanwhile learning intent-aware user/item representations. Knowledge Contrastive Denoising under Intents is dedicated to learning precise and robust representations. It leverages intent-aware representations to sample relevant knowledge, and proposes a local-global contrastive mechanism to enhance noise-irrelevant representation learning. Extensive experiments conducted on benchmark datasets show the superior performance of our proposed method over the state-of-the-arts. And online A/B testing results on Alibaba large-scale industrial recommendation platform also indicate the real-scenario effectiveness of KGTN.

cs.IR↗
arXiv2024

Search Intenion Network for Personalized Query Auto-Completion in E-Commerce

Query Auto-Completion(QAC), as an important part of the modern search engine, plays a key role in complementing user queries and helping them refine their search intentions.Today's QAC systems in real-world scenarios face two major challenges:1)intention equivocality(IE): during the user's typing process,the prefix often contains a combination of characters and subwords, which makes the current intention ambiguous and difficult to model.2)intention transfer (IT):previous works make personalized recommendations based on users' historical sequences, but ignore the search intention transfer.However, the current intention extracted from prefix may be contrary to the historical preferences.

cs.IR↗
arXiv2023

Gaussian Graph with Prototypical Contrastive Learning in E-Commerce Bundle Recommendation

Bundle recommendation aims to provide a bundle of items to satisfy the user preference on e-commerce platform. Existing successful solutions are based on the contrastive graph learning paradigm where graph neural networks (GNNs) are employed to learn representations from user-level and bundle-level graph views with a contrastive learning module to enhance the cooperative association between different views. Nevertheless, they ignore the uncertainty issue which has a significant impact in real bundle recommendation scenarios due to the lack of discriminative information caused by highly sparsity or diversity. We further suggest that their instancewise contrastive learning fails to distinguish the semantically similar negatives (i.e., sampling bias issue), resulting in performance degradation. In this paper, we propose a novel Gaussian Graph with Prototypical Contrastive Learning (GPCL) framework to overcome these challenges. In particular, GPCL embeds each user/bundle/item as a Gaussian distribution rather than a fixed vector. We further design a prototypical contrastive learning module to capture the contextual information and mitigate the sampling bias issue. Extensive experiments demonstrate that benefiting from the proposed components, we achieve new state-of-the-art performance compared to previous methods on several public datasets. Moreover, GPCL has been deployed on real-world e-commerce platform and achieved substantial improvements.

cs.IR↗
arXiv2023

Transformer-based Language Model Fine-tuning Methods for COVID-19 Fake News Detection

With the pandemic of COVID-19, relevant fake news is spreading all over the sky throughout the social media. Believing in them without discrimination can cause great trouble to people's life. However, universal language models may perform weakly in these fake news detection for lack of large-scale annotated data and sufficient semantic understanding of domain-specific knowledge. While the model trained on corresponding corpora is also mediocre for insufficient learning. In this paper, we propose a novel transformer-based language model fine-tuning approach for these fake news detection. First, the token vocabulary of individual model is expanded for the actual semantics of professional phrases. Second, we adapt the heated-up softmax loss to distinguish the hard-mining samples, which are common for fake news because of the disambiguation of short text. Then, we involve adversarial training to improve the model's robustness. Last, the predicted features extracted by universal language model RoBERTa and domain-specific model CT-BERT are fused by one multiple layer perception to integrate fine-grained and high-level specific representations. Quantitative experimental results evaluated on existing COVID-19 fake news dataset show its superior performances compared to the state-of-the-art methods among various evaluation metrics. Furthermore, the best weighted average F1 score achieves 99.02%.

cs.CL↗
arXiv2022

Trustworthy Knowledge Graph Completion Based on Multi-sourced Noisy Data

Knowledge graphs (KGs) have become a valuable asset for many AI applications. Although some KGs contain plenty of facts, they are widely acknowledged as incomplete. To address this issue, many KG completion methods are proposed. Among them, open KG completion methods leverage the Web to find missing facts. However, noisy data collected from diverse sources may damage the completion accuracy. In this paper, we propose a new trustworthy method that exploits facts for a KG based on multi-sourced noisy data and existing facts in the KG. Specifically, we introduce a graph neural network with a holistic scoring function to judge the plausibility of facts with various value types. We design value alignment networks to resolve the heterogeneity between values and map them to entities even outside the KG. Furthermore, we present a truth inference model that incorporates data source qualities into the fact scoring function, and design a semi-supervised learning way to infer the truths from heterogeneous values. We conduct extensive experiments to compare our method with the state-of-the-arts. The results show that our method achieves superior accuracy not only in completing missing facts but also in discovering new facts.

cs.IR↗
arXiv2021

Metapaths guided Neighbors aggregated Network for?Heterogeneous Graph Reasoning

Most real-world datasets are inherently heterogeneous graphs, which involve a diversity of node and relation types. Heterogeneous graph embedding is to learn the structure and semantic information from the graph, and then embed it into the low-dimensional node representation. Existing methods usually capture the composite relation of a heterogeneous graph by defining metapath, which represent a semantic of the graph. However, these methods either ignore node attributes, or discard the local and global information of the graph, or only consider one metapath. To address these limitations, we propose a Metapaths-guided Neighbors-aggregated Heterogeneous Graph Neural Network(MHN) to improve performance. Specially, MHN employs node base embedding to encapsulate node attributes, BFS and DFS neighbors aggregation within a metapath to capture local and global information, and metapaths aggregation to combine different semantics of the heterogeneous graph. We conduct extensive experiments for the proposed MHN on three real-world heterogeneous graph datasets, including node classification, link prediction and online A/B test on Alibaba mobile application. Results demonstrate that MHN performs better than other state-of-the-art baselines.

cs.AI↗
arXiv2020

Spending Money Wisely: Online Electronic Coupon Allocation based on Real-Time User Intent Detection

Online electronic coupon (e-coupon) is becoming a primary tool for e-commerce platforms to attract users to place orders. E-coupons are the digital equivalent of traditional paper coupons which provide customers with discounts or gifts. One of the fundamental problems related is how to deliver e-coupons with minimal cost while users' willingness to place an order is maximized. We call this problem the coupon allocation problem. This is a non-trivial problem since the number of regular users on a mature e-platform often reaches hundreds of millions and the types of e-coupons to be allocated are often multiple. The policy space is extremely large and the online allocation has to satisfy a budget constraint. Besides, one can never observe the responses of one user under different policies which increases the uncertainty of the policy making process. Previous work fails to deal with these challenges. In this paper, we decompose the coupon allocation task into two subtasks: the user intent detection task and the allocation task. Accordingly, we propose a two-stage solution: at the first stage (detection stage), we put forward a novel Instantaneous Intent Detection Network (IIDN) which takes the user-coupon features as input and predicts user real-time intents; at the second stage (allocation stage), we model the allocation problem as a Multiple-Choice Knapsack Problem (MCKP) and provide a computational efficient allocation method using the intents predicted at the detection stage. We conduct extensive online and offline experiments and the results show the superiority of our proposed framework, which has brought great profits to the platform and continues to function online.

cs.AI↗