arXiv subjects

Wenjun Yang

Publications and source records attributed to Wenjun Yang.

12 recordsLinked to original sources

arXiv2026

DREAM Technical Report

Industrial recommender systems commonly use cascaded retrieval, ranking, and re-ranking pipelines. Although efficient, these pipelines fragment information and objectives across modules, rely on rigid rules, and have limited awareness of real-time intent, leaving session-level shifts among browsing, comparison, and purchase insufficiently addressed. We present DREAM (Developing Recommender Engine with Agentic Methods), an autonomous optimization control architecture that adds a perception-aware, orchestrable, and auditable policy layer atop existing pipelines without replacing them. DREAM has two core components. First, a three-tier Intent Engine fuses on-device signals into structured L0/L1/L2 intent representations; its edge-cloud trigger chain reduces reporting volume to approximately 8.7%. Second, a Meta Engine uses a MetaModel for layered M1-to-M2-to-M3 reasoning: intent summarization, strategy planning informed by Strategy Memory, and parameter translation. It dispatches the resulting parameters through a unified outlet with safety guardrails. A Reward Dual Loop continuously optimizes both components by combining offline simulation for strategy-space exploration with online feedback for outcome calibration, forming a cycle of generation, execution, evaluation, and experience accumulation. Large-scale A/B tests on Taobao's homepage feed show that re-ranking control alone improves IPV by 2.06%, Core IPV by 2.39%, and GMV by 0.88%. Extending control to fine ranking raises these gains to 2.71%, 3.06%, and 1.31%, respectively, while consistently improving PV by more than 1%. These gains require neither replacement of pipeline models nor compromise of serving stability, supporting agentic meta-control as a viable paradigm for industrial recommendation.

cs.IR↗

arXiv2026

RecGPT-V3 Technical Report

Large language models (LLMs) are transforming recommender systems from matching co-occurrence patterns in historical behavior toward reasoning about the intent that drives it. RecGPT-V1 pioneered this paradigm on Taobao by centering user understanding, and RecGPT-V2 scaled it via coordinated multi-agent reasoning; both are deployed in production with consistent gains in user experience and commercial outcomes. However, operating RecGPT at scale reveals three challenges: (1) stateless behavior modeling, where each request reprocesses full user history, wasting computation and discarding prior analysis; (2) a tag-to-item information bottleneck, where natural-language tags form a lossy channel between user understanding and item grounding; and (3) inefficient explicit reasoning, whose lengthy chain-of-thought incurs untenable latency and compute overhead. We present RecGPT-V3, a stateful, hybrid-modal recommender that reasons over natural language for open-world knowledge and Semantic IDs (SIDs) for concrete item grounding. A Memory Hub maintains structured, continually evolving user memory that distills long-horizon behavior into condensed units, cutting user-modeling computation by 55.8%. A Hybrid-modal Foundation Model allows the LLM jointly reason over text tags and SIDs, opening a high-bandwidth channel into the item space. Latent Intent Reasoning internalizes verbose rationales into compact learnable latent tokens that remain decodable into readable explanations, lowering output token cost by 200x. Deployed in Taobao's "Guess What You Like" feed, RecGPT-V3 achieves consistent gains in large-scale online A/B tests: IPV +1.28%, CTR +1.00%, TC +1.97%, GMV +3.97%, while cutting end-to-end serving resource consumption by 52.4%.

cs.IR↗

arXiv2025

RecGPT-V2 Technical Report

Large language models (LLMs) have demonstrated remarkable potential in transforming recommender systems from implicit behavioral pattern matching to explicit intent reasoning. While RecGPT-V1 successfully pioneered this paradigm by integrating LLM-based reasoning into user interest mining and item tag prediction, it suffers from four fundamental limitations: (1) computational inefficiency and cognitive redundancy across multiple reasoning routes; (2) insufficient explanation diversity in fixed-template generation; (3) limited generalization under supervised learning paradigms; and (4) simplistic outcome-focused evaluation that fails to match human standards. To address these challenges, we present RecGPT-V2 with four key innovations. First, a Hierarchical Multi-Agent System restructures intent reasoning through coordinated collaboration, eliminating cognitive duplication while enabling diverse intent coverage. Combined with Hybrid Representation Inference that compresses user-behavior contexts, our framework reduces GPU consumption by 60% and improves exclusive recall from 9.39% to 10.99%. Second, a Meta-Prompting framework dynamically generates contextually adaptive prompts, improving explanation diversity by +7.3%. Third, constrained reinforcement learning mitigates multi-reward conflicts, achieving +24.1% improvement in tag prediction and +13.0% in explanation acceptance. Fourth, an Agent-as-a-Judge framework decomposes assessment into multi-step reasoning, improving human preference alignment. Online A/B tests on Taobao demonstrate significant improvements: +2.98% CTR, +3.71% IPV, +2.19% TV, and +11.46% NER. RecGPT-V2 establishes both the technical feasibility and commercial viability of deploying LLM-powered intent reasoning at scale, bridging the gap between cognitive exploration and industrial utility.

cs.IR↗

arXiv2025

BIER-Star: Stateless Geographic Multicast for Scalable Satellite-Terrestrial Integration

The rapid expansion of LEO satellite constellations has enabled an integrated terrestrial network and non-terrestrial network (TN-NTN), connecting diverse users such as aircraft, ships, and remote communities. These networks increasingly need a scalable and efficient multicast protocol for critical applications like emergency alerts, large-scale software updates, and real-time broadcasting. However, traditional multicast protocols, such as IP-based multicast and software-defined multicast approaches, introduce significant control overhead and struggle to adapt to the dynamic and mobile nature of satellite topologies. This paper presents BIER-Star, a stateless multicast protocol designed for the integrated TN-NTN. BIER-Star uses a two-layer geospatial gridding scheme (i.e., H3) to encode destinations as Earth- and space-cell identifiers rather than per-terminal addresses. This cell-based abstraction shortens the header bitstring, simplifies forwarding, and eliminates per-flow state and complex signaling. Our simulations indicate that BIER-Star reduces header size versus BIER and avoids geographic path-finding failures seen in greedy methods.

cs.NI↗

arXiv2025

Interpreting Convolutional Neural Network Activation Maps with Hand-crafted Radiomics Features on Progression of Pediatric Craniopharyngioma after Irradiation Therapy

Purpose: Convolutional neural networks (CNNs) are promising in predicting treatment outcome for pediatric craniopharyngioma while the decision mechanisms are difficult to interpret. We compared the activation maps of CNN with hand crafted radiomics features of a densely connected artificial neural network (ANN) to correlate with clinical decisions. Methods: A cohort of 100 pediatric craniopharyngioma patients were included. Binary tumor progression was classified by an ANN and CNN with input of T1w, T2w, and FLAIR MRI. Hand-crafted radiomic features were calculated from the MRI using the LifeX software and key features were selected by Group lasso regularization, comparing to the activation maps of CNN. We evaluated the radiomics models by accuracy, area under receiver operational curve (AUC), and confusion matrices. Results: The average accuracy of T1w, T2w, and FLAIR MRI was 0.85, 0.92, and 0.86 (ANOVA, F = 1.96, P = 0.18) with ANN; 0.83, 0.81, and 0.70 (ANOVA, F = 10.11, P = 0.003) with CNN. The average AUC of ANN was 0.91, 0.97, and 0.90; 0.86, 0.88, and 0.75 of CNN for the 3 MRI, respectively. The activation maps were correlated with tumor shape, min and max intensity, and texture features. Conclusions: The tumor progression for pediatric patients with craniopharyngioma achieved promising accuracy with ANN and CNN model. The activation maps extracted from different levels were interpreted with hand-crafted key features of ANN.

physics.med-ph↗

arXiv2025

Deep-learning-based Radiomics on Mitigating Post-treatment Obesity for Pediatric Craniopharyngioma Patients after Surgery and Proton Therapy

Purpose: We developed an artificial neural network (ANN) combining radiomics with clinical and dosimetric features to predict the extent of body mass index (BMI) increase after surgery and proton therapy, with advantage of improved accuracy and integrated key feature selection. Methods and Materials: Uniform treatment protocol composing of limited surgery and proton radiotherapy was given to 84 pediatric craniopharyngioma patients (aged 1-20 years). Post-treatment obesity was classified into 3 groups (<10%, 10-20%, and >20%) based on the normalized BMI increase during a 5-year follow-up. We developed a densely connected 4-layer ANN with radiomics calculated from pre-surgery MRI (T1w, T2w, and FLAIR), combining clinical and dosimetric features as input. Accuracy, area under operative curve (AUC), and confusion matrices were compared with random forest (RF) models in a 5-fold cross-validation. The Group lasso regularization optimized a sparse connection to input neurons to identify key features from high-dimensional input. Results: Classification accuracy of the ANN reached above 0.9 for T1w, T2w, and FLAIR MRI. Confusion matrices showed high true positive rates of above 0.9 while the false positive rates were below 0.2. Approximately 10 key features selected for T1w, T2w, and FLAIR MRI, respectively. The ANN improved classification accuracy by 10% or 5% when compared to RF models without or with radiomic features. Conclusion: The ANN model improved classification accuracy on post-treatment obesity compared to conventional statistics models. The clinical features selected by Group lasso regularization confirmed our practical observation, while the additional radiomic and dosimetric features could serve as imaging markers and mitigation methods on post-treatment obesity for pediatric craniopharyngioma patients.

physics.med-ph↗

arXiv2025

RecGPT Technical Report

Recommender systems are among the most impactful applications of artificial intelligence, serving as critical infrastructure connecting users, merchants, and platforms. However, most current industrial systems remain heavily reliant on historical co-occurrence patterns and log-fitting objectives, i.e., optimizing for past user interactions without explicitly modeling user intent. This log-fitting approach often leads to overfitting to narrow historical preferences, failing to capture users' evolving and latent interests. As a result, it reinforces filter bubbles and long-tail phenomena, ultimately harming user experience and threatening the sustainability of the whole recommendation ecosystem. To address these challenges, we rethink the overall design paradigm of recommender systems and propose RecGPT, a next-generation framework that places user intent at the center of the recommendation pipeline. By integrating large language models (LLMs) into key stages of user interest mining, item retrieval, and explanation generation, RecGPT transforms log-fitting recommendation into an intent-centric process. To effectively align general-purpose LLMs to the above domain-specific recommendation tasks at scale, RecGPT incorporates a multi-stage training paradigm, which integrates reasoning-enhanced pre-alignment and self-training evolution, guided by a Human-LLM cooperative judge system. Currently, RecGPT has been fully deployed on the Taobao App. Online experiments demonstrate that RecGPT achieves consistent performance gains across stakeholders: users benefit from increased content diversity and satisfaction, merchants and the platform gain greater exposure and conversions. These comprehensive improvement results across all stakeholders validates that LLM-driven, intent-centric design can foster a more sustainable and mutually beneficial recommendation ecosystem.

cs.IR↗

arXiv2025

KORGym: A Dynamic Game Platform for LLM Reasoning Evaluation

Recent advancements in large language models (LLMs) underscore the need for more comprehensive evaluation methods to accurately assess their reasoning capabilities. Existing benchmarks are often domain-specific and thus cannot fully capture an LLM's general reasoning potential. To address this limitation, we introduce the Knowledge Orthogonal Reasoning Gymnasium (KORGym), a dynamic evaluation platform inspired by KOR-Bench and Gymnasium. KORGym offers over fifty games in either textual or visual formats and supports interactive, multi-turn assessments with reinforcement learning scenarios. Using KORGym, we conduct extensive experiments on 19 LLMs and 8 VLMs, revealing consistent reasoning patterns within model families and demonstrating the superior performance of closed-source models. Further analysis examines the effects of modality, reasoning strategies, reinforcement learning techniques, and response length on model performance. We expect KORGym to become a valuable resource for advancing LLM reasoning research and developing evaluation methodologies suited to complex, interactive environments.

cs.CL↗

arXiv2025

Democratizing Differential Privacy: A Participatory AI Framework for Public Decision-Making

This paper introduces a conversational interface system that enables participatory design of differentially private AI systems in public sector applications. Addressing the challenge of balancing mathematical privacy guarantees with democratic accountability, we propose three key contributions: (1) an adaptive $\epsilon$-selection protocol leveraging TOPSIS multi-criteria decision analysis to align citizen preferences with differential privacy (DP) parameters, (2) an explainable noise-injection framework featuring real-time Mean Absolute Error (MAE) visualizations and GPT-4-powered impact analysis, and (3) an integrated legal-compliance mechanism that dynamically modulates privacy budgets based on evolving regulatory constraints. Our results advance participatory AI practices by demonstrating how conversational interfaces can enhance public engagement in algorithmic privacy mechanisms, ensuring that privacy-preserving AI in public sector governance remains both mathematically robust and democratically accountable.

cs.IT↗

arXiv2022

Overview of the Observing System and Initial Scientific Accomplishments of the East Asian VLBI Network (EAVN)

The East Asian VLBI Network (EAVN) is an international VLBI facility in East Asia and is operated under mutual collaboration between East Asian countries, as well as part of Southeast Asian and European countries. EAVN currently consists of 16 radio telescopes and three correlators located in China, Japan, and Korea, and is operated mainly at three frequency bands, 6.7, 22, and 43 GHz with the longest baseline length of 5078 km, resulting in the highest angular resolution of 0.28 milliarcseconds at 43 GHz. One of distinct capabilities of EAVN is multi-frequency simultaneous data reception at nine telescopes, which enable us to employ the frequency phase transfer technique to obtain better sensitivity at higher observing frequencies. EAVN started its open-use program in the second half of 2018, providing a total observing time of more than 1100 hours in a year. EAVN fills geographical gap in global VLBI array, resulting in enabling us to conduct contiguous high-resolution VLBI observations. EAVN has produced various scientific accomplishments especially in observations toward active galactic nuclei, evolved stars, and star-forming regions. These activities motivate us to initiate launch of the 'Global VLBI Alliance' to provide an opportunity of VLBI observation with the longest baselines on the earth.

astro-ph.IM↗

arXiv2017

VLBI ecliptic plane survey: VEPS-1

We present here the results of the first part of the VLBI Ecliptic Plane Survey (VEPS) program. The goal of the program is to find all compact sources within $7.5^\circ$ of the ecliptic plane which are suitable as calibrators for anticipated phase referencing observations of spacecraft and determine their positions with accuracy at the 1.5~nrad level. We run the program in two modes: the search mode and the refining mode. In the search mode, a complete sample of all sources brighter than 50 mJy at 5 GHz listed in the Parkes-MIT-NRAO (PMN) and Green Bank 6~cm (GB6) catalogs, except those previously detected with VLBI, is observed. In the refining mode, the positions of all ecliptic plane sources, including those found in the search mode, are improved. By October 2016, thirteen 24-hr sessions that targeted all sources brighter than 100~mJy have been observed and analyzed. Among 3320 observed target sources, 555 objects have been detected. We also conducted a number of follow-up VLBI experiments in the refining mode and improved the positions of 249 ecliptic plane sources.

astro-ph.IM↗

arXiv2016

Planetary Radio Interferometry and Doppler Experiment (PRIDE) technique: A test case of the Mars Express Phobos fly-by

The closest ever fly-by of the Martian moon Phobos, performed by the European Space Agency's Mars Express spacecraft, gives a unique opportunity to sharpen and test the Planetary Radio Interferometry and Doppler Experiments (PRIDE) technique in the interest of studying planet - satellite systems. The aim of this work is to demonstrate a technique of providing high precision positional and Doppler measurements of planetary spacecraft using the Mars Express spacecraft. The technique will be used in the framework of Planetary Radio Interferometry and Doppler Experiments in various planetary missions, in particular in fly-by mode. We advanced a novel approach to spacecraft data processing using the techniques of Doppler and phase-referenced very long baseline interferometry spacecraft tracking. We achieved, on average, mHz precision (30 {\mu}m/s at a 10 seconds integration time) for radial three-way Doppler estimates and sub-nanoradian precision for lateral position measurements, which in a linear measure (at a distance of 1.4 AU) corresponds to ~50 m.

astro-ph.IM↗