Alibaba’s advertising team introduces semantic‑aware contrastive learning to pre‑train multimodal image‑text embeddings, integrates them via SimTier and MAKE into ID‑based CTR models, achieving up to 6.9% lift in Taobao display ad click‑through rates and improving long‑tail item performance.
Alibaba’s AI team introduced AtomoVideo, a diffusion‑based image‑to‑video generator enhanced by a training‑free Noise Rectification module that adds and corrects controlled noise to eliminate first‑frame errors, enabling merchants to automatically create high‑fidelity 4‑second 720p product videos with strong temporal consistency for e‑commerce advertising.
This paper proposes a two‑stage Potential Generalized Second Price auction for augmented ads, ranking guide ads by expected welfare from their linked second‑step ads, shifting billing to the second click to eliminate free‑riding, and demonstrates via offline and online experiments on Taobao that it boosts click‑through, revenue, and GMV while lowering CPC.
The Xiaohongshu anti‑fraud team combats sophisticated same‑group and crowdsourced reselling bots by ingesting real‑time transaction streams into a Nebula Graph, using multi‑hop sub‑graph sampling, label propagation, and modularity‑based community detection to identify suspicious clusters, update risk pools, and enforce personalized purchase‑limit rules.
The article details a high‑concurrency e‑commerce coupon system that separates creation and distribution, uses Redis‑Lua atomic stock deduction with asynchronous replenishment and coupon generation, employs a lightweight distributed‑transaction table, and adds bucket and batch optimizations to safely handle tens of thousands of TPS.
The article details vivo’s transition from a monolithic mall to a micro‑service transaction platform, outlining a multi‑tenant architecture with ShardingSphere‑sharded MySQL, Snowflake IDs, Elasticsearch search, configurable state machines, generic delayed tasks, Seata and local‑message distributed transactions, plus high‑availability safeguards, emphasizing pragmatic solution selection.
The article details how an e‑commerce product middle platform’s front‑end employs visualization technology, a uni‑render approach that shares a Vuex store between an iframe and its parent via a sandboxed Vue instance, and a qiankun‑based micro‑frontend architecture, while addressing cross‑domain and editor integration challenges.
The Vivo online mall introduces a multi‑dimensional pricing capability matrix and floor‑price monitoring system that flags or blocks orders when combined promotions push SKU prices below predefined thresholds, providing pre‑event configuration checks, real‑time alerts during sales, and post‑event analysis to prevent loss and price‑hunting incidents.
The article details Vivo Mall’s new promotion pricing engine, which replaces duplicated site‑level logic with a layered model (product, shop, platform) that supports configurable stacking, processes discounts through three calculation stages, and uses a meta‑driven rule engine to flexibly interpret JSON promotion templates, ensuring extensible, stable pricing functionality.
The article chronicles Vivo’s official mall transformation from a 2015 monolithic Java MVC system to a service‑oriented architecture with independent activity, product, coupon, and order services, later expanding to consignment, CPS, and promotion subsystems, and finally adding multi‑language, multi‑timezone, and multi‑region capabilities for global markets.
YouZan built a side‑car, real‑time loss‑prevention system that monitors business events and database binlogs, validates information flow and fund balance, enforces idempotency and manual‑change checks, and uses refund‑settlement‑cash‑out circuit‑breakers to automatically detect and stop financial loss, achieving successful fault prevention after six months.
The article details Baidu’s 2020‑built commodity promotion system for Baijiahao and live‑stream e‑commerce, linking merchants with authors/streamers through CPS billing, three user interfaces and five core services, and highlights technical choices such as dynamic‑library tracking, asynchronous batch writes, and a high‑cohesion, low‑coupling architecture requiring cross‑team collaboration.
The paper presents an intelligent reach system for e‑commerce that automatically selects audience, timing, channel, welfare and creative content using user, content and decision models—including XGBoost churn predictions, NLP‑generated copy, Bayesian CTR estimation and linear‑programming optimization—resulting in a 17.4 % rise in paying users and a 5 % revenue boost over manual methods.
The article details the design and implementation of a dedicated pricing calculation engine that decouples complex e‑commerce promotion logic from transaction services, using componentized rules, configurable pipelines, multi‑level caching, and lane‑based simulations to achieve high extensibility, low latency, and accurate pricing across browsing and checkout scenarios.
The paper presents CAIG, a CTR‑driven advertising image generation pipeline that pre‑trains a multimodal LLM on e‑commerce data, trains a reward model on CTR‑labeled image pairs, and fine‑tunes generation via product‑centric preference optimization, achieving state‑of‑the‑art online and offline performance.
Tech Insight highlights ten 2024 JD Retail Technology AI papers presented at top conferences—including CVPR, SIGIR, WWW, AAAI and IJCAI—that advance open‑vocabulary object detection, unified search‑recommendation, pre‑ranking consistency, diversity‑aware re‑ranking, a diversified product‑search dataset, graph‑based query classification, plug‑in CTR models, parallel ad‑ranking, trajectory‑based CTR stability, and task‑aware decoding for large language models.
This article recounts the development of JD.Vision as a native Apple Vision Pro application, detailing how spatial computing reshapes e‑commerce experiences, the challenges of home‑scene product visualization, user‑type strategies, and the product‑manager growth lessons drawn from the project.
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 presents a comprehensive overview of JD Retail's AI‑driven multi‑agent architecture for e‑commerce assistance, detailing how real‑world merchant decision processes are modeled with ReAct‑based LLM agents, the hierarchical workflow, training pipelines, monitoring mechanisms, and future directions for scalable intelligent commerce support.
The article analyzes JD.com's recommendation advertising ranking auction mechanism, detailing its objectives, challenges in traffic value estimation, user interest exploration, and multi‑item auction fairness, and describing the technical evolution from traditional auctions to deep‑learning‑driven solutions.