This article details Toutiao’s rapid growth and its large‑scale data pipeline, covering article crawling, user modeling, recommendation engines, storage solutions, push notifications, micro‑service architecture, and the underlying virtualization PaaS platform that powers its personalized news service.
This article introduces DyHAN, a hierarchical attention‑based dynamic heterogeneous graph embedding method that captures temporal, node‑level, and edge‑level information, demonstrates superior offline and online performance on Alibaba’s ICBU recommendation system, and discusses dataset construction, model architecture, and future challenges.
The article surveys typical challenges in recommendation systems, including ambiguous evaluation metrics, the trade‑off between precise algorithms and user experience, the exploration‑exploitation dilemma, and why offline AUC improvements often lead to online CTR/CPM drops due to data leakage, feature inconsistency, and distribution shifts.
This article details how Toutiao processes billions of daily page views, builds user models with Hadoop and Storm, runs real‑time recommendation and cold‑start personalization, and scales its microservice‑based architecture using Kafka, MySQL, MongoDB, Redis and a high‑throughput push system.
This article outlines the engineering architecture of a fine‑grained, personalized recommendation system for e‑commerce, covering core components such as feature data (offline and real‑time), algorithm engine (recall and ranking), technology choices like MongoDB, Elasticsearch, Kafka, Redis, and model deployment strategies.
This article describes how 58 Tongzhen leverages AI technologies—including data pipelines, feature engineering, various recall and ranking models, and AB‑testing—to build a personalized feed recommendation system for the down‑market, detailing its overall architecture, data sources, model iterations, performance gains, and future directions.
This article presents the overall design, multi‑dimensional control mechanisms, auxiliary modules, and evolutionary processes of the quality control system used in WeChat's Look feature, detailing data lifecycle, model training, generalization, transfer learning, and continuous anti‑abuse strategies.
This article explains how Huajiao Live designed and evolved its live‑streaming recommendation system, covering basic concepts, collaborative‑filtering and matrix‑factorization techniques, deep‑learning models, ranking and multi‑objective optimization, and practical deployment considerations for real‑time personalized content delivery.
This article explains the cold‑start problem in recommendation systems, outlines common industry solutions such as popular‑content, group‑representative, auxiliary‑information, bandit algorithms, and deep learning, and details how Huajiao Live applied these techniques to improve new‑user engagement and metrics.
This article details how Alibaba's 1688 B2B platform leverages deep learning techniques—including Deep Match, DIN, DIEN, DMR, and heterogeneous network models—to evolve its product recall, ranking, and live‑content recommendation pipelines, highlighting system architecture, practical lessons, and online performance improvements.
This article presents a comprehensive technical overview of Meitu's personalized push notification pipeline, detailing the evolution of embedding methods (Word2Vec, Airbnb listing embedding, graph embedding), multiple recall strategies (global, personalized, attribute, and content‑based), and a progression of ranking models from logistic regression to field‑wise three‑tower architectures, highlighting their impact on click‑through rates.
This article uses a vivid kite‑selling story to illustrate how user profiling, data tagging, and recommendation tactics can be combined to increase transaction volume, improve average order value, and avoid common pitfalls such as unclear goals, poor data quality, and unvalidated tags.
This article presents a comprehensive overview of Phoenix News's AI‑driven feed recommendation system, detailing its business challenges, multi‑stage architecture, deep learning models, feature pipelines, metric trade‑offs, cold‑start solutions, and practical insights for improving user satisfaction and content quality.
The article presents the Deep Match to Rank (DMR) model, which integrates collaborative‑filtering inspired user‑to‑item relevance modeling into the ranking stage of recommendation systems, achieving significant offline and online improvements in click‑through rate and revenue metrics for e‑commerce platforms.
This article details the architecture, model evolution, feature processing, online learning, and future optimization plans of AutoHome's recommendation system, covering stages from resource collection to ranking, various models such as LR, XGBoost, FM, DeepFM, and operational practices like AB testing and debugging.
This article describes the end‑to‑end recommendation pipeline of the Province Money Fast Report app, covering business background, data collection, model training and evaluation, the evolution from FM to DeepFM, DIN, DCN, xDeepFM, ESMM and custom networks, as well as serving strategies and practical lessons learned.
The talk explains how Beike Real Estate constructs a large‑scale relationship graph from billions of user, house, and agent interactions, quantifies edge strengths, builds homogeneous and heterogeneous sub‑graphs, derives graph capabilities such as node influence, embedding, similarity and relation prediction, and finally deploys these capabilities in multi‑degree queries, house‑similarity recommendations and B‑side agent assistance, achieving measurable CTR improvements.
This article details Toutiao’s large‑scale recommendation system, explaining how it models content, user, and environment features, the variety of algorithms and real‑time training pipelines used, feature engineering categories, recall strategies, content analysis, user tagging, evaluation methods, and content‑safety mechanisms.
This article lists several 2019 whitepapers from Sensor Data covering how enterprises can build recommendation systems from scratch, construct tag‑profile frameworks, integrate data research with product development, implement enterprise event‑tracking methodologies, and understand big‑data technologies and architectures.
The article outlines a micro‑service based intelligent recommendation platform that supports over 40 scenarios, detailing its overall architecture, AB testing service, and the three core modules—index, recall, and filter—while also describing future plans for platform centralization and open development.
This article examines the limitations of cloud‑based recommendation pipelines, explains how edge computing can provide localized user perception and rapid re‑ranking, describes the EdgeRec on‑device model architecture—including heterogeneous behavior sequence modeling and behavior‑aware attention reranking—and presents offline and online experimental results that demonstrate significant gains in click‑through and conversion rates.
This article provides a comprehensive overview of Toutiao's rapid growth and technical architecture, detailing its massive user base, data collection pipelines, user modeling, recommendation engines, storage solutions, message push mechanisms, micro‑service design, and virtualization PaaS platform.
Alibaba’s Taobao team applied large‑scale edge computing and deep neural networks on the client side to continuously sense user intent, enabling real‑time, personalized recommendations that significantly boosted click‑through rates and GMV during major shopping events.
Effective short‑video recommendation relies on meticulous feature engineering that transforms raw signals—numerical counts, categorical IDs, content and user embeddings, context and session data—through bucketization, scaling, crossing, and smoothing, then selects and evaluates them via filtering, wrapping, regularization, and importance analysis to mitigate business biases and improve multi‑objective ranking performance.
When the recommendation service of the Mafengwo app experiences database disconnections, third‑party timeouts, or network jitter, a locally‑deployed off‑heap cache built with OHC and SpringBoot can return pre‑computed results, isolating business logic, reducing latency, and improving user experience during failures.
This article details how 58.com constructed a large‑scale personalized recommendation platform for its recruitment business, covering business background, user intent modeling, knowledge‑graph and NER techniques, user profiling, multi‑stage recall strategies, ranking model pipelines, serving infrastructure, AB testing, and future research directions.
This article explores how Alibaba’s 3D scene‑based recommendation system combines computer‑vision, deep‑learning and data‑driven matching algorithms to create immersive, size‑accurate product visualizations that enhance user experience and drive higher click‑through rates in e‑commerce.
This article explains how MaFengWo’s recommendation system combines recall, ranking, and rerank stages, details the evolution of its sorting algorithm platform, and shows how data verification and model‑explainability techniques like SHAP and LIME improve online performance and accelerate model iteration.
The article details JD.com's internally developed distributed search engine and recommendation platform, their new architectures, deep‑learning‑driven ranking and recall models, component‑based deployment, extensive performance testing, and coordinated operations that powered the massive 11.11 shopping event.
This article describes a standardized workflow for feature collection, configuration, processing, and model training/prediction in large‑scale recommendation systems, using CSV‑based definitions and code generation to ensure consistency between offline training and online serving while reducing manual coding effort.
iQIYI’s feed recommendation system adopts an online‑learning framework that continuously trains a massive Wide‑and‑Deep DNN on billions of streaming samples, handling dynamic user interests, OOV embeddings, delayed labels, and non‑convex optimization, enabling hourly model refreshes and delivering up to 3.8 % higher consumption versus offline baselines.
This article describes the challenges of large‑scale model training for Snowball’s feed recommendation, and details a series of engineering optimizations—including GPU acceleration, multi‑threaded data preparation, TFRecord conversion, compression, and batch‑map reordering—that increased training throughput from 6 k to over 20 k samples per second while reducing CPU and I/O bottlenecks.
This article presents a comprehensive overview of short‑video recommendation at UC, covering business background, system architecture, the evolution from LR to Wide & Deep models, multi‑objective loss design with positive‑sample weighting, graph‑embedding fusion, time‑weighted loss, continuity modeling, a Boosting‑based WnD solution, and future research directions.
This article explains how Alibaba’s Hand‑Taobao platform uses a full‑link intelligent delivery framework—combining user intent recognition, rights recommendation, and advanced machine‑learning models such as XFTRL and Thompson Sampling—to predict activity drops during the Spring Festival and deliver personalized interventions that significantly improve DAU, click‑through, and redemption rates.
This article explains the architecture and principles of Toutiao’s recommendation system, covering its three‑dimensional model of content, user and environment features, content analysis techniques, user tagging, real‑time training pipelines, evaluation methods, and content safety measures that together drive personalized feeds.
This article details Kuaikan Comics' personalized recommendation pipeline, covering business context, diverse content formats, technical challenges, content‑based and collaborative‑filtering methods, ranking models, system architecture, A/B testing, and future directions for improving recommendation quality.
This article provides a comprehensive introduction to recommendation systems, covering their definition, overall offline and online architectures, feature engineering, collaborative filtering, latent semantic models, ranking algorithms, and evaluation methods including A/B testing and offline metrics.
This article details how the MX recommendation team rebuilt their Faiss‑based similarity search service using gRPC, covering service selection, multi‑type vector handling, dynamic index updates, deployment strategies, and performance gains that doubled QPS and cut latency by two‑thirds.
This article examines Taobao’s homepage recommendation system, comparing the full‑page browsing mode that displays a continuous stream of items with the section‑focused mode that groups products into distinct blocks, highlighting their design principles, user experience implications, and underlying recommendation strategies.
This article explains MX Player’s recommendation indexing architecture, detailing batch and incremental build processes, real‑time and Falcon monitoring, Redis storage design, and how these components together improve system stability, reduce latency, and enhance user experience.
WeChat Reading’s “Guess You Like” engine combines real‑time click and reading data with tag‑based and deep‑learning embeddings, LSH similarity search, and a DeepFM ranking model to deliver cross‑type book, article, and video recommendations, continuously balancing exploitation and exploration to boost CTR and user engagement.
The Real‑time Attention‑based Look‑alike Model (RALM) converts recommendation to a user‑user problem by representing items with aggregated seed‑user embeddings, employs shared projection, local and global attention towers, and enables instant, diverse, high‑CTR recommendations without retraining, as demonstrated by its deployment in WeChat “Look‑at”.
This article explains MX Player’s recommendation engine architecture, detailing the division between functional and algorithmic components, API communication via Thrift, and the configurable core processing pipeline that handles recall, filtering, scoring, ranking, rule‑based adjustments, packaging, and caching.
This article shares extensive practical experience on recommendation system optimization, outlining the importance of problem definition, the limits of AB testing, and four guiding principles—avoid fundamentally wrong actions, do the right things correctly, keep solutions simple, and prevent over‑optimization.
The article outlines JD’s evolution from rule‑based product suggestions in 2012 to a sophisticated, AI‑driven, multi‑screen personalized recommendation platform, detailing its product types, system architecture, data collection, offline and online computation, and the core recommendation engine that powers features like “Guess You Like.”
This article traces the evolution of Weibo's recommendation system architecture across three major stages—independent 1.0, layered 2.0, and platform‑style 3.0—detailing the environmental drivers, technical goals, component composition, advantages, shortcomings, and the concrete outcomes of each phase.
This article presents a comprehensive overview of Alibaba Retail's B2B2C intelligent recommendation platform, detailing its business context, three core recommendation scenarios, system architecture, matching algorithms such as item‑CF, graph embedding and user‑CF, as well as the evolution of ranking models and feature engineering practices.
This article explains how Alibaba's App recommendation pipeline integrates marketing scenario cards using weak personalization and machine‑learning models, detailing the metrics, feature engineering, recall and ranking strategies that together raise exposure revenue and click‑through performance.
This article explores the multifaceted criteria for evaluating a good recommendation system, covering macro and micro perspectives, product domain considerations, information retrieval, algorithmic accuracy, user experience, and business impact, and outlines a systematic iteration process for continuous improvement.
This article presents an overview of Ctrip's travel recommendation system, detailing its architecture, user‑behavior analysis, product catalog handling, various recall strategies, ranking methods—including machine‑learning models like XGBoost—and future directions toward deeper AI and NLP integration.
The article summarizes the 58 Group technical salon where experts presented the microservice‑based recommendation system architecture, data and strategy layers, and the internally built search ranking platform covering sampling, feature engineering, and model training, highlighting practical implementations and lessons learned.
When the recommendation service experiences database disconnections, third‑party timeouts, or network jitter, we designed an off‑heap local disaster‑recovery cache using OHC and SpringBoot that isolates cache logic, writes asynchronously, backs up to disk, and restores availability, keeping latency under 10 ms and improving user experience.
This article explains how to build high‑performance, highly‑available, scalable, extensible, and secure recommendation services by outlining background concepts, defining quality criteria, discussing design challenges, and presenting concrete architectural principles and practical strategies.
This article explains the design of a live‑streaming recommendation system, covering its overall architecture, ranking, content‑based and collaborative‑filtering methods, similarity calculations, multi‑algorithm fusion, sorting, user profiling, and evaluation metrics with practical examples and diagrams.
This article presents a comprehensive overview of personalized recommendation systems, covering their purpose, common algorithms, development challenges, the multi‑layer architecture used at DataGrand, optimization techniques, and the range of services offered to enterprise customers.
The NVID recommendation system for Taobao is built on a four‑layer architecture—activity material, configuration, business process, and application—and solves environment isolation, performance, audience management, and A/B testing challenges through optimized data schemas, ID mapping, multi‑level caching with database fallback, and real‑time user targeting, while future work aims at personalized audiences and automated ad optimization.
The article explains content‑based filtering for recommendation systems, covering its basic concepts, feature requirements, implementation using vector representations and cosine similarity, advantages and disadvantages, and supplementary algorithms such as k‑Nearest Neighbor, Rocchio, decision trees, linear classifiers, and Naive Bayes.
This article provides a comprehensive overview of Toutiao’s recommendation system, detailing its three‑dimensional modeling of content, user, and context, the feature extraction pipeline, real‑time training infrastructure, user‑tag generation, evaluation methodology, and content‑safety mechanisms.
This article details the end‑to‑end design, recall and ranking techniques, engineering implementation, and future research directions of Tencent's Yoo video bottom‑page recommendation system, illustrating how large‑scale video recommendation is built from business needs to deep learning models.
This article explains the design of a Cross‑Domain Embedding (CSDE) method that transfers Alipay user features to Taobao representations, details its learning and adaptive prediction stages, and shows experimental and online results demonstrating significant conversion‑rate improvements for new and inactive users.
The talk details how Beike's senior algorithm expert Xu Yansong designed, iterated, and scaled a multi‑stage intelligent recommendation platform for real‑estate, covering service upgrades, personalized algorithms such as collaborative filtering and user profiling, modular architecture, stability engineering, data feedback loops, and future AI‑driven enhancements.
This article details how 51 Credit Card leverages artificial intelligence to build a personalized recommendation system, covering business pain points, technical challenges, a three‑layer tagging architecture from bill and app data, model deployment pipelines, and real‑world growth case studies that boosted conversion and ROI.
The article summarizes a QCon 2018 presentation by Luo Jingjia of Manbang Group, detailing how AI and deep‑learning techniques are applied to model, optimize, and recommend vehicle‑cargo matching in China’s massive highway logistics network, including data handling, neural‑network design, and practical challenges.
This article describes Tubi's experiment with a Long Short‑Term Memory (LSTM) based recommender, covering the model architecture, training setup, performance metrics, and observed recommendation improvements on their video streaming platform.
This article explains how e‑commerce recommendation systems work, covering regular and personalized recommendation types, the challenges of user profiling and data handling, the three‑stage recommendation pipeline, and the overall system architecture that powers real‑time, AI‑driven product suggestions.
This article details the design, architecture, and key components of 360 Mobile Assistant's recommendation system, covering business scenarios, data warehouse and computing layers, feature engineering, model selection, and online deployment strategies to improve app discovery and user engagement.
At KDD 2018 in London, JD presented ten research papers—including a reinforcement‑learning based recommendation system called DEERS—highlighting its big‑data and AI capabilities across retail, supply‑chain, healthcare, and smart‑city applications while fostering industry‑academic collaboration.
Meipai’s personalized recommendation system leverages massive user‑behavior data to build behavior‑driven video clusters—evolving from TopicModel through Item2vec and Keyword Propagation to a DSSM deep model—boosting ranking AUC, enhancing UI diversity, similar‑video search, niche discovery, and feature engineering.
The talk covered Meipai’s personalized deep ranking model that balances depth and low latency, a behavior‑driven video clustering method that enriches recommendation beyond visual cues, and the use of advanced data structures to accelerate solving large‑scale optimization problems in business contexts.
This article explains the online side of recommendation systems, covering data flow, traffic allocation, AB testing, the gateway, recall mechanisms, and real-time ranking, while also describing how offline-trained models are deployed as binary weight files for live serving.
This article explains the goals, architectures, data pipelines, recall strategies, and ranking models of contemporary recommendation systems, covering both online and offline components, collaborative filtering, content-based methods, feature engineering, and practical interview insights for engineers.
This article explains the fundamentals and advanced techniques of e‑commerce product recommendation systems, covering conventional and personalized approaches, user profiling, data collection, storage, modeling, the three‑stage pipeline of preprocessing, recall and ranking, as well as system architecture, challenges, and key algorithms such as LR and GBDT.
This article details JD's evolution from rule‑based recommendations to a multi‑screen, AI‑driven personalization platform, describing its system architecture, data pipelines, feature services, and key technologies that enable real‑time, user‑centric product suggestions across the e‑commerce ecosystem.
The article explains the core concepts of recommendation engines—user profiling and data classification—describes how large‑scale data processing tools are used to build models, and outlines common offline and A/B testing approaches for evaluating recommendation performance.
This article details how 58.com leverages deep learning models such as FNN, Wide&Deep, CNN+DNN, and YouTube DNN recall, along with a custom AI platform, to enhance recommendation ranking and recall, achieving measurable improvements in click‑through rates and overall system performance.
Alibaba’s latest bundle‑mining system leverages weighted graph embedding and real‑time sampling to recommend complementary products, replacing traditional item‑to‑item similarity, boosting click‑through rates by up to 13% offline and 4% online during the Double‑11 promotion while handling billions of edges.
This article presents a comprehensive overview of Toutiao's recommendation system, detailing its three‑dimensional modeling approach, real‑time training pipeline, feature engineering, content and user analysis techniques, evaluation methodology, and the extensive content‑safety mechanisms employed to ensure reliable and responsible information distribution.
This article provides a comprehensive overview of Toutiao's recommendation system, covering its three‑dimensional modeling approach, feature engineering, real‑time training pipeline, recall strategies, user‑tag generation, evaluation methodology, and content‑safety mechanisms.
This article explains how Alibaba's recommendation team built a large‑scale, AI‑driven personalization pipeline for the Double 11 shopping festival, using graph‑embedding recall, deep‑learning ranking models such as DeepResNet, DCN, and a custom XTensorflow platform to improve coverage, diversity, and click‑through rates.
The article details Alibaba's Taobao Personalization Platform (TPP) intelligent scheduling system, explaining its architecture, optimization algorithms, convergence logic, and performance results that dramatically improve CPU utilization and automate scaling during both regular operation and high‑traffic events like Double‑11.
This tutorial explains various similarity metrics such as cosine similarity, Euclidean distance, Jaccard index, and Pearson correlation, demonstrates a Python function to compute user interest similarity, and shows how to generate movie recommendations with example code and output.
This tutorial walks you through the fundamentals of recommendation systems, explaining why they’re needed, the types of engines, collaborative filtering concepts, and provides a hands‑on Python implementation with sample data and code to build your own recommender from the ground up.
The sixth Meitu Internet Technology Salon in Beijing showcased Meipai’s evolution, with senior engineers detailing the platform’s recommendation system, real‑time background segmentation, monitoring framework, live‑streaming and bullet‑screen architecture, offering practical insights and best‑practice lessons for building and optimizing large‑scale video services.
This article details the architecture and evolution of a headline recommendation system, covering data aggregation, storage strategies using MySQL and Redis, challenges with reload latency and memory usage, and the optimizations—including data separation, Redis migration, and query pipeline improvements—that enabled scalable, efficient backend operations.
The article describes the design, storage, update mechanisms, and optimization strategies of a headline recommendation platform that aggregates various data types using algorithms, MySQL, Redis, and a modular data‑fetching framework to achieve scalable and efficient content delivery.
The paper describes Meituan‑Dianping’s two‑stage recommendation pipeline—recall and ranking—and how a Wide & Deep neural architecture, enriched with extensive user, item, and context features and trained with Adam and cross‑entropy loss, significantly boosts CTR and recommendation novelty, with future plans to add RNNs and reinforcement learning.
This article explains the background, four‑layer architecture, offline data mining techniques, and various user‑behavior‑based methods—including interaction, similar‑user, and collaborative‑filtering approaches—used to build accurate Weibo post recommendation systems.
This article examines Toutiao’s rapid rise as a personalized news platform, detailing its AI‑driven recommendation pipeline, web‑crawling infrastructure, similarity‑matrix algorithms, A/B testing, and the role of human moderation in delivering highly targeted content to billions of users.
The article details how Ctrip’s CTO Gan Quan has leveraged big‑data platforms, deep‑learning algorithms, cross‑screen user tracking, and rapid AB testing to create a real‑time, personalized recommendation engine that shortens travel decision cycles and drives significant revenue growth.
This article recounts Wu Guanlin’s presentation on the evolution of Vipshop’s personalized recommendation system, detailing the technical challenges of real‑time predictions, the three generations of architecture, the four‑stage recommendation engine, and the VRE platform’s design for scalability and low latency.
The Youku Tudou data team overhauled its video recommendation system by moving ranking from offline to online, detailing architectural changes, advantages, challenges, feature handling, offline evaluation, and model weight fusion to improve scalability and user experience.
Alibaba’s merchant division introduced a scalable graph‑embedding approach for its “thousands‑of‑people‑one‑face” recommendation module, enabling personalized product suggestions within sparse shop data, improving click‑through rates by 30% and conversions by 60%, and presenting theoretical insights validated at AAAI 2017.
This article explains how Tencent built a billion‑scale, generalized recommendation system by designing a reusable algorithm library, deploying a low‑latency, highly available real‑time streaming platform (R2), and offering a cloud‑based recommendation engine that simplifies integration for internet businesses.
This article outlines a service‑oriented architecture for a high‑availability personalized recommendation platform, detailing the front‑end, back‑end, crawler, user‑profile modeling, data collection from logs and client events, and processing pipelines using technologies such as Node.js, Python, RabbitMQ/Kafka, MongoDB and TensorFlow.
This article explains how comprehensive user profiling—combining static demographics and dynamic behavior logs—feeds recommendation engines, detailing data sources, feature extraction, ranking formulas, and the long‑term goals of delivering personalized, high‑quality content to users.
This article describes how Ctrip transformed its traditional application architecture into a high‑concurrency, big‑data‑driven platform, detailing storage, compute, and business‑layer redesigns that enable massive data ingestion, real‑time user‑intent services, and a scalable personalized recommendation system.
This article outlines Toutiao's rapid growth, its data pipelines, user modeling, recommendation system, storage solutions, and push notification strategies that together enable personalized news delivery to hundreds of millions of users.
This article details Meituan Dianping's data‑driven approach to precise marketing, describing the O2O marketing framework, a layered pyramid data system, profiling techniques, budget monitoring, and two real‑world case studies that together illustrate how big‑data technologies boost marketing efficiency on mobile platforms.
This article presents Meituan Dianping's data‑driven precise marketing architecture, detailing a layered pyramid system, user profiling, budget monitoring, and two real‑world cases—potential user mining and a smart coupon engine—demonstrating how big‑data techniques improve marketing efficiency and ROI.
This article explains the architecture of both system-wide and personalized recommendation modules, compares three recommendation strategies, details the use of Apache Mahout for collaborative filtering with Java code examples, and discusses cold‑start solutions within a Spark‑Hadoop stack.
