The Xianyu team built a data‑driven system that monitors seller online status and reply speed, uses Siddhi CEP to match behavior patterns, and orchestrates activities, tasks, and synchronization modules, boosting conversion by three percentage points and allowing new scenarios to launch without developer effort.
Laser (Latent Surrogate Representation learning) estimates long‑term ride‑hailing market effects by inferring hidden surrogate variables from short‑term outcomes using an iVAE and inverse‑probability weighting, thereby reducing experiment cost and latency while achieving more accurate causal effect predictions than existing baselines.
Online click‑through‑rate (CTR) prediction demands millisecond‑level response times, yet deep models are too slow; this paper introduces a “Rocket Launching” framework that jointly trains a lightweight net and a powerful booster net, sharing parameters and using gradient‑blocking and hint loss to improve accuracy without increasing inference latency.
The talk, led by Bilibili’s risk‑control head, outlines Internet risk‑control fundamentals, precise traffic perception techniques, and a full‑scenario joint‑defense framework that combines hierarchical identification, cross‑scene signal sharing, statistical anomaly detection, and layered mitigation (soft and hard) to counter black‑market attacks on platforms.
This article outlines a HiMCM modeling problem that asks high school students to identify factors, build a decision algorithm, test it with fictional cases, and propose a presentation format for selecting the best summer job, balancing earnings, recreation, and personal preferences.
Xiaohongshu’s commercial team modeled the real‑time bidding process for its “Fries” ad product, derived an optimal linear‑programming bid formula, and implemented a simple two‑parameter PID‑controlled scheme that meets client pacing, delivery guarantees, and platform profit goals while using practical heuristics.
This article presents a comprehensive design for Zhuanzhuan's home‑page recommendation pipeline, detailing the system architecture, challenges of traffic efficiency and diversity, and a two‑stage solution that applies Proximal Policy Optimization reinforcement learning in the re‑ranking module and Determinantal Point Process optimization in the coarse‑ranking and traffic‑pool stages, followed by offline simulation, online deployment, and evaluation metrics.
Rocket Launching introduces a novel co‑training framework that jointly trains a lightweight network and a more powerful booster network, sharing parameters and using gradient‑blocking and hint loss to improve click‑through‑rate prediction accuracy while keeping online inference latency unchanged, validated on public datasets and Alibaba’s ad system.
The article presents the Universal Recommendation Model (URM), a large‑language‑model‑based recall framework that integrates world knowledge and e‑commerce expertise through knowledge injection and prompt‑driven alignment, achieving significant offline recall gains and a 3.1% increase in ad consumption while meeting high‑QPS, low‑latency production constraints.
This article explains the concept of referral (or distribution) business, compares e‑commerce and home‑service referral models, and details the design of a flexible, node‑based commission architecture—including rule configuration, strategy‑pattern implementation, and workflow automation—to support multi‑stage payouts in service‑oriented platforms.
JD's advertising research team combines deep learning, multimodal modeling, reinforcement‑learning auctions, and generative recommendation to boost ad relevance, improve long‑tail product exposure, and overcome large‑model inference challenges in a high‑traffic e‑commerce environment.
This article explores how mathematical modeling and graph‑theoretic methods can rigorously solve three classic Afanti folk stories—dividing donkeys, splitting oil, and a meal‑fee dispute—illustrating the power of formal models for everyday problem solving.
This article introduces PODM‑MI, a preference‑oriented diversity model that uses mutual information and variational Gaussian representations to jointly optimize accuracy and diversity in e‑commerce search re‑ranking, and reports significant online A/B test improvements on JD.com.
This article reviews eight Kuaishou‑authored papers accepted at ICLR 2026, summarizing their problem statements, novel methods such as front‑door causal attribution, visual table retrieval, denoising rerankers, difficulty‑adaptive reasoning, diffusion code infilling, generative ordinal regression, multimodal video retrieval, e‑commerce dialogue benchmarks, and a new LLM creativity evaluator, together with reported experimental gains.
MaRCA, a multi‑agent reinforcement‑learning framework, allocates compute across JD’s advertising playback chain by jointly estimating user value, resource consumption, and action outcomes while dynamically adjusting to real‑time load, achieving roughly 15 % higher ad revenue without extra compute resources.
The article presents a comprehensive anti‑fraud framework that analyzes black‑market fraud characteristics, reviews conventional fraud‑mitigation methods, and proposes a multimodal graph‑neural‑network approach—leveraging device, behavior, and content similarity—to accurately identify fraudulent users on large‑scale internet platforms.
The 2021 Taobao Push project introduced an AI‑driven copy‑generation platform that combines template extraction and fine‑tuned Unilm models with diverse beam search, creating diverse, high‑quality push messages, cutting manual costs, and delivering a 10 % click‑through lift and higher material adoption.
The Sustainable Online Reinforcement Learning (SORL) framework tackles offline inconsistency in auto‑bidding by iteratively gathering safe online data from real ad systems with a Lipschitz‑based exploration method and training a variance‑suppressed conservative Q‑learning policy, achieving safer, more stable, and higher‑performing bids on Alibaba’s platform.
The article presents Alibaba's real‑time computing platform for search, detailing its micro‑ and macro‑level architectures, online learning frameworks, point‑wise and pair‑wise ranking models, bandit‑based strategy optimization, and PID‑controlled traffic regulation, and reports significant performance gains during the Double‑11 shopping festival.
The article highlights ten representative Meituan technical papers accepted at ACM SIGIR 2022, spanning personalized opinion tagging, cross‑domain sentiment classification, dialogue summarization transfer, universal retrieval, CTR prediction, image behavior modeling, and topic segmentation, each summarized with abstracts and download links for researchers.
