This article explains Huya Live's recommendation architecture, covering business background, system design, vector retrieval challenges and solutions with ScaNN, and the fine‑ranking pipeline, while highlighting performance optimizations, scalability, and future directions for their live‑streaming platform.
This article explains how Kuaishou detects and predicts viral short‑video素材 by defining content types, outlining essential viral elements, describing a two‑stage coarse‑recall and fine‑ranking model that combines speed‑based features, Gaussian mixture modeling, and a lightweight DNN, and showcases real‑world case studies and Q&A.
This article walks through the development of a simple advertising recommendation system using Python, covering data collection, preprocessing with label encoding, text embedding via Torch, constructing an MLP model, and initiating training, while reflecting on the challenges faced by Python developers in the big‑data era.
The Dewu app’s new multi‑objective ranking model replaces the shallow ESMM baseline with a DeepFM‑based MLP and a Deep Interest Transformer that encodes up to 120 recent user actions, adds a dedicated bias network, and fuses short‑ and long‑term interests, achieving modest CTR and CVR AUC improvements while planning future tab‑specific extensions.
The seventh Scala Meetup gathered over 1400 online participants to share three technical talks covering Tubi's content recommendation system architecture, philosophical insights into the nature of computation, and practical experiences with reactive streams in large‑scale JVM environments, followed by a round‑table discussion and audience feedback.
This article details the evolution of the ‘逛逛’ content community’s recommendation system, comparing the legacy rule‑based Hive workflow with a new algorithm‑driven architecture that leverages Elasticsearch, Redis, multi‑stage recall, coarse‑ and fine‑ranking, re‑ranking, exposure filtering, cold‑start handling, performance tuning, and future plans for vector‑based recall and platformization.
This article presents a comprehensive overview of Home's intelligent recommendation system, detailing its business value, challenges of the previous siloed approach, the platform‑based architecture with standardized capabilities, micro‑service components, and the observability stack that together enable scalable, personalized content delivery for millions of users.
The article details Guangguang’s shift from a rule‑based, Hive‑driven recommendation pipeline to an algorithmic service that leverages Elasticsearch and Redis for multi‑source recall, coarse and fine model ranking, exposure filtering, cold‑start handling, latency optimizations, reliability monitoring, and future vector‑based enhancements.
This talk explains how Meituan builds and leverages a massive lifestyle-domain knowledge graph to improve LBS recommendation, covering explicit and implicit graph applications, challenges such as explainability and data sparsity, and advanced models like dual‑memory networks and cross‑domain learning.
This article examines the practical aspects of open‑sourcing Twitter’s recommendation algorithm, covering the platform’s data model, timeline views, ranking features, a TypeScript pseudocode illustration, and the major engineering challenges of scale, real‑time processing, reliability, and security.
This article presents Kuaishou's large‑scale recommendation system, detailing its precision‑ranking architecture, the PPNet CTR model, multi‑domain multi‑task learning, short‑ and long‑term behavior sequence modeling, billion‑feature trillion‑parameter engineering, and future research directions.
This article presents Meituan's large‑scale knowledge graph, its integration into location‑based recommendation, the challenges of explainability, domain diversity, data sparsity and spatiotemporal complexity, and describes a dual‑memory neural network and cross‑domain learning approach that improve recall, ranking and recommendation fairness.
This article details a year‑long series of experiments on the Taobao “Every Square Every House” infinite‑scroll light app, describing how added recall paths, a coarse‑ranking filter, multi‑task MMOE sorting, a lightweight down‑scroll predictor, and relevance‑enhanced features together boosted click‑through, scroll depth and per‑user engagement by double‑digit percentages.
This article explains the background, design principles, high‑concurrency handling, storage optimization, rule‑engine implementation, and intelligent scheduling needed to create a universal, stable, extensible, and intelligent recommendation slot that enriches short‑video app ecosystems.
The article presents NetEase Cloud Music’s tag‑based cold‑start recommendation approach, which generates song tags, computes similarity, and recommends new, unevaluated tracks to users, addressing data scarcity and quality challenges while improving exposure coverage and balancing user experience, establishing a vital foundation for a healthy music ecosystem.
The article presents Snowball's model feature platform, detailing its motivation, architecture, feature lifecycle management, online engine design, optimization techniques, and the resulting improvements in feature iteration speed, reuse, and system stability for recommendation and search services.
This article presents Kuaishou's graph‑based item‑to‑item (I2I) recall pipeline for short‑video recommendation, detailing the business challenges, pipeline architecture, optimization techniques such as similarity‑measure tricks, graph structure learning, edge‑weight learning, and future research directions.
This article explains how MaFengWo's recommendation ranking system tackles position bias by incorporating position features, using inverse propensity weighting, and adjusting click metrics, resulting in measurable improvements in click‑through rate, content exposure, and overall recommendation accuracy.
The article details Alibaba’s end‑to‑end online‑learning pipeline for real‑time ranking in the Taobao home‑decor channel, covering UT log parsing, full‑feature extraction, ODL sample creation, xDeepCTR model training, and deployment, which yielded up to 7.8% CTR improvement and demonstrates the value of rapid model adaptation.
This article details how QQ Music tackled rapid growth in recommendation traffic by redesigning its recommendation architecture, introducing a cloud‑native machine‑learning platform, optimizing data services, and adopting a DAG‑based recall system to improve scalability, flexibility, and development efficiency.
The article provides a thorough guide to ranking in recommendation systems, detailing the pipeline architecture, sample handling challenges, extensive feature engineering categories, the evolution from collaborative filtering to deep and attention‑based models, and key optimization trade‑offs between memorization, generalization, and efficient user‑interest modeling.
This article presents a comprehensive overview of Huya Live's recommendation system, covering business background, system architecture, vector retrieval techniques, ranking pipeline, technical challenges, implementation details, and future outlook, highlighting scalability and performance optimizations.
The article details how DaJia's recommendation system progressed through three architectural versions—V1.0's simple strategy‑factory design, V2.0's vertical business split and configurable pipeline, and V3.0's dynamic configuration service and modular pipeline—addressing scalability, fault isolation, and personalization challenges while outlining future directions for explainable AI recommendations.
This article details the evolution of an e‑commerce recommendation system through three architectural versions, highlighting the initial simple design, the challenges that prompted vertical and horizontal splits, the introduction of a configurable pipeline and AB testing, and the final micro‑service‑based, dynamically configurable V3.0 architecture.
This article details the technical evolution of Kuaishou's short‑video recommendation pipeline, focusing on sequence re‑ranking, multi‑content mixing, and on‑device re‑ranking, and explains how transformer‑based models, generator‑evaluator frameworks, and reinforcement‑learning strategies are employed to maximize overall sequence value, user engagement, and revenue.
This article details the three‑stage evolution of an e‑commerce recommendation framework—V1.0’s simple strategy‑factory design, V2.0’s vertical business split, and V3.0’s configurable pipeline with dynamic server‑client configuration, addressing scalability, isolation, and AB‑testing challenges.
This article details the evolution of the "到家" recommendation system across three major versions, describing the initial strategy‑factory prototype, the problems it faced, the vertical and horizontal splits introduced in V2.0, the dynamic pipeline configuration and service separation in V3.0, and the future outlook for finer‑grained personalization.
This article examines the step‑by‑step evolution of the recommendation framework used by a major e‑commerce service, detailing the shortcomings of the initial V1.0 design, the vertical and modular refinements introduced in V2.0, and the dynamic configuration, pipeline, and service‑oriented enhancements implemented in V3.0 to improve scalability, stability, and fine‑grained experimentation.
This article presents a comprehensive overview of Feizhu's information‑flow recommendation system, detailing its mixed‑material architecture, cold‑start recall and coarse‑ranking techniques, multi‑modal pre‑training and fine‑tuning, fine‑ranking with user‑state gating, and tiered traffic‑flow mechanisms for content delivery.
A's fruit‑ordering platform kept adding tags, preferences, and fallback strategies to please every user, but the growing complexity made it hard to configure and debug, ultimately driving customers back to B's straightforward manual ordering system.
DeWu Community’s recommendation system progressed from single‑objective CTR modeling to a multi‑objective framework that combines independent models for dwell time, video completion and user interactions via score‑fusion, ranking‑learning and multi‑task architectures with shared parameters and gradient‑blocking, delivering higher engagement and retention.
This article explains why recommendation systems are essential for B2B construction e‑commerce, describes the types of data they rely on, outlines multi‑channel recall methods, details collaborative‑filtering algorithms with similarity calculations, and presents the four‑stage recommendation pipeline from recall to re‑ranking.
This talk explains how 58.com’s massive blue‑collar recruitment platform uses reinforcement‑learning techniques—including multi‑armed bandits, contextual MAB, and linear UCB—to address cold‑start and interest‑divergence challenges, describes the system architecture, offline evaluation, online deployment, and reports an 8% uplift in new‑user conversion.
This talk presents the design and implementation of 58's commercial recruitment recommendation system, covering the business scenario, system architecture, regional and behavior‑based recall methods, various ranking models—including coarse‑ranking, dual‑tower, DIN‑bias, and multitask W3DA—and future optimization directions.
This talk presents the design and implementation of 58’s commercial recruitment recommendation system, covering the characteristics of the app’s recommendation scenario, system architecture, region‑based and behavior‑based recall methods, and coarse‑ and fine‑ranking models with various optimizations and future directions.
This article presents a comprehensive overview of the 58.com commercial recruitment recommendation system, detailing its business challenges, system architecture, region‑based and behavior‑based recall strategies, coarse‑ and fine‑ranking models, bias handling, evaluation methods, and future directions.
iQIYI’s short‑video team solves the new‑video cold‑start problem by using a GAN that generates latent user features from video attributes and a discriminator to validate them, then matches these vectors to real users via cosine similarity, achieving double‑digit gains in exposure, CTR, and watch time.
This article presents QQ Music’s end‑to‑end solution for data‑driven content understanding, value evaluation, and fine‑grained operation, detailing offline and real‑time pipelines, neural‑network models, a content middle‑platform, parameter services, and a precise delivery system that boost user engagement while preserving experience.
iQIYI’s generic ranking framework unifies feature production, replay, training, and ranking into modular, configurable phases that handle offline and real‑time data, support diverse models, provide automated monitoring, and have been deployed across all platforms, delivering over 20% higher watch time and doubling first‑play videos.
The article presents a Guard abstraction that temporarily borrows and returns Protobuf field pointers via set_allocated/release, eliminating costly CopyFrom operations in recommendation pipelines, enabling zero‑copy field sharing across central‑control and recall stages, improving CPU usage and latency while handling safety and rollback.
Meituan’s To‑Store Integrated data team built an end‑to‑end supply‑standardization pipeline for the rapidly growing script‑murder market by extending the GENE knowledge graph to mine merchant supply, construct a unified script library through rule‑based, semantic, and multimodal clustering, and link products and user‑generated content to standardized scripts, enabling a dedicated category, personalized recommendations, filter tags, and improved ranking.
CONDE, a concept‑aware denoising graph neural network proposed by Wuhan University and Kuaishou, leverages heterogeneous three‑part graphs, attention‑based graph convolutions, and Gumbel‑Softmax‑driven edge sampling to filter noisy user‑video interactions, achieving up to 6 % AUC improvement on short‑video and e‑commerce recommendation tasks.
Baidu’s massive information‑flow recommendation platform employs an intelligent delivery pipeline—spanning micro‑service R&D, automated white‑box testing, performance monitoring, and optimized deployment—that supports nearly a hundred daily releases, cuts QA effort, delivers over half of requests within a day, and enables near‑zero‑touch, high‑frequency rollouts.
The 58.com AI algorithm competition showcased over 210 teams competing to improve job recommendation click‑through and conversion rates, featured an award ceremony with speeches, highlighted the ten winning teams, and presented detailed solution shares—including tree models, feature‑engineering techniques, and deep‑learning approaches—while offering GPU resources on the WPAI platform for continued participation.
This article presents the end‑to‑end practice of a second‑hand housing recommendation system at 58.com and Anjuke, covering business background, embedding‑based vector recall, multi‑objective ranking methods such as ESMM and MMOE, experimental results, and future development directions.
This article explains the architecture and key components of Toutiao’s recommendation system, covering system overview, content analysis, user tagging, evaluation methods, and content safety measures, and discusses practical implementation details such as feature engineering, model training, recall strategies, and online experimentation.
The article details Haolo's end-to-end AI platform—built on Hadoop/Yarn with Spark ML, XGBoost and TensorFlow—and explains how its matching recommendation engine, transaction-ecosystem governance models, and intelligent uplift-based marketing system jointly boost carpool efficiency, safety, user retention, and ROI.
This article examines the rapid growth of short‑video platforms, outlines the core problems a recommendation system must solve for users, creators, and advertisers, describes the end‑to‑end pipeline, explores multi‑objective modeling and fusion techniques, and discusses long‑term value estimation for sustained user engagement.
iQIYI’s engineering team built an online vector‑recall service on Milvus, wrapping it with a Dubbo‑gRPC interface to serve 6 M 64‑dimensional embeddings at roughly 3 k QPS and 20 ms p99 latency, integrating query‑embedding generation, simplifying recommendation pipelines, and demonstrating the performance and operational advantages of a platformized ANN‑based recall layer.
Gorse is an open‑source Go recommendation system that automates model selection, supports distributed training and prediction, offers a RESTful API and dashboard, and stores data in MySQL/MongoDB with Redis caching, enabling fast integration of personalized suggestions into online services.
Gorse is an open‑source, Go‑based recommendation system that automates model selection, supports distributed training and real‑time serving via RESTful APIs, stores data in MySQL or MongoDB with Redis caching, and offers a dashboard for monitoring, data import/export, and cluster health.
This article explains how a two‑stage recommendation pipeline uses explicit tag‑based recall, inverted indexes, and a multi‑task learning model to improve click‑through and dwell time by dynamically balancing loss weights across tasks.
This article describes how Baidu engineered a flexible, multi‑level fault‑tolerant architecture—including dynamic retry scheduling, multi‑recall coordination, ranking layer degradation, and cross‑IDC multi‑master storage—to achieve five‑nine availability for its massive feed recommendation service.
This article explains how to design and implement a lightweight Python recommendation system for personal use by replacing rating‑based similarity with label‑based scoring, handling sparse data, and dynamically updating user models with time‑limited label weights.
This article presents an in‑depth overview of Tencent Music's live‑streaming recommendation system, covering business background, system architecture, recall and ranking model designs, multi‑modal extensions, and advanced training techniques such as DSSM, ESMM, GradNorm, and CGC to improve user engagement and conversion.
This article presents a comprehensive overview of Haola's ride‑sharing algorithm ecosystem, covering the machine‑learning platform foundation, the architecture and evolution of the matching recommendation engine, the transaction‑ecosystem governance models, and the intelligent marketing uplift framework, highlighting technical challenges, solutions, and performance gains.
iQIYI improves short‑video recommendation by applying multi‑objective optimization—weighting clicks by watch duration, fusing separate click and watch‑time models, employing multi‑task learning with ESMM/MMOE and Pareto‑guided PSO hyper‑parameter search—delivering 7%+ watch‑time growth, 20%+ interaction gains, and 1.5‑3% CTR lifts while planning cross‑scene learning and further model refinements.
This article presents the technical exploration of splicing recall in flight ticket search at Alibaba's Fliggy, detailing the background, challenges, constrained routing and machine‑learning algorithms, the four‑step solution pipeline, and the resulting improvements in ticket availability and conversion rates.
This article presents HRL-Rec, a hierarchical reinforcement learning model that jointly learns user preferences at the item and channel levels for integrated recommendation systems, and demonstrates its superior offline and online performance, stability, and scalability through extensive experiments on the WeChat "See" platform.
This article walks through a complete real‑time recommendation system built on Apache Flink, detailing its v2.0 architecture, modules for user behavior, interest, and product profiling, the recommendation algorithms (hot‑list, collaborative filtering, item similarity), and step‑by‑step Docker deployment of MySQL, Redis, HBase, and Kafka.
This article walks through a complete Flink‑based recommendation system, detailing its v2.0 architecture, recommendation algorithms, front‑end and back‑end components, and step‑by‑step Docker deployment of MySQL, Redis, HBase, and Kafka services.
This article presents a comprehensive walkthrough of a Flink‑powered recommendation system, detailing its v2.0 architecture, module functions, recommendation algorithms (hotness, product similarity, collaborative filtering), front‑end and back‑end UI, and step‑by‑step Docker deployment of MySQL, Redis, HBase, and Kafka services.
This article presents a comprehensive overview of a Flink-powered real-time recommendation system, detailing its v2.0 architecture, module functions, recommendation algorithms, front‑end and back‑end interfaces, Docker‑based deployment of MySQL, Redis, HBase, Kafka, and step‑by‑step startup procedures.
The article provides a comprehensive technical overview of Toutiao's rapid growth, detailing its massive user base, data collection and processing pipelines, user modeling, cold‑start strategies, recommendation engines, storage solutions, push notification mechanisms, and the underlying microservice and PaaS architecture.
This article details how 58.com improved its home‑page recommendation system by introducing vectorized recall with Word2Vec, optimizing negative sampling, deploying FAISS for fast nearest‑neighbor search, and later adopting a dual‑tower deep learning model with user interest features, achieving higher click‑through and conversion rates.
This article provides a comprehensive overview of Toutiao's rapid growth, massive user base, data collection pipelines, user modeling, recommendation engine, storage solutions, message push strategies, micro‑service architecture, and virtualization PaaS platform, illustrating how big‑data technologies enable personalized news delivery at scale.
This article presents a comprehensive technical overview of the K‑Song recommendation platform, covering its backend architecture, the evolution of recall strategies, feature management and ranking pipelines, large‑scale deduplication techniques, and the debugging and monitoring infrastructure that support high‑performance personalized music recommendations.
This article details Baidu's Feed recommendation system architecture, explaining how a combination of dynamic retry scheduling, real‑time stop‑loss mechanisms, multi‑recall frameworks, ranking layer fallbacks, and IDC‑level multi‑master designs collectively ensure five‑nine availability across billions of daily requests.
The Vivo App Store recommendation system uses a modular, hot‑plug architecture—layered from plugins to base services and employing Template Method, Strategy, and Composite patterns—to integrate unchanged data sources, deliver high‑performance personalized recommendations, reduce development effort by ~75%, and enable rapid, low‑bug scene‑specific upgrades.
This article details the business background, system architecture, coarse‑ranking algorithms, dual‑tower model, model distillation, diversity‑control techniques such as DPP, and the online performance gains of the recommendation pipeline used in Tencent Music's Quanmin K‑Song platform.
This article presents the design and implementation of a multi‑objective optimization framework using Multi‑gate Mixture‑of‑Experts (MMoE) to improve click‑through, conversion, and purchase behaviors in Taobao's "Lying Flat" home‑goods recommendation channel, detailing model variants, feature engineering, loss weighting, and online A/B test results.
This article details the practical upgrades of iQIYI's short‑video recommendation coarse‑ranking pipeline, moving from a GBDT model to a dual‑tower DNN, applying knowledge distillation, embedding compression, inference optimizations, and finally a cascade architecture to align with the fine‑ranking model while reducing resource consumption.
iQIYI’s short‑video recommendation team replaced a GBDT coarse‑ranking model with a lightweight dual‑tower DNN, applied knowledge distillation, sparse‑aware embedding optimization, and inference merging, then introduced a cascade MMOE architecture, achieving comparable accuracy with half the memory, ~19 ms latency reduction, and measurable gains in watch time, CTR and engagement.
This article presents EdgeRec, a comprehensive edge‑computing framework for recommendation systems that redesigns the architecture, introduces on‑device real‑time user perception, proposes a context‑aware reranking model (CRBAN), details on‑device mixing and training pipelines, and demonstrates significant business improvements through extensive experiments and deployments.
This article presents the end‑to‑end design, feature engineering, model evolution (Wide&Deep, DeepFM, DCN, DIN, DIEN), multi‑task training, and deployment optimizations that 58.com applied to improve search ranking for its rental business, demonstrating significant gains in click‑through and conversion rates.
This article describes how a community’s big‑data team designed a content hotness algorithm by defining time, interaction, content, and user dimensions, assigning business meanings, applying weighted formulas and a Newton‑cooling decay function, and integrating user interest vectors to compute dynamic scores.
This article presents a comprehensive overview of a recommendation system, covering its business background, purpose, detailed engineering architecture—including data sources, computation, storage, online learning, service and access layers—and discusses key challenges, module design, and practical reflections.
This article explains the fundamentals of recommendation systems, covering data collection, user and content profiling, system architecture, algorithmic pipelines such as recall, filtering, ranking, and evaluation metrics, while also discussing practical challenges like echo chambers and long‑term user value.
This case study details 58.com's full‑time recruitment page redesign, shifting from group‑based to individual‑focused personalization, introducing three recommendation zones, a feedback loop, and inclusive, fun UI elements to boost user engagement and application rates.
This article details the evolution of Beike's commercial strategy algorithm platform, describing its business scenarios, bidding mechanisms, architecture redesign across online, near‑real‑time, and offline layers, model training, vector retrieval, service governance, and the performance and stability improvements achieved.
The paper presents a WeChat recommendation system that estimates user interest tags via multi‑class classification, using hierarchical intra‑ and inter‑domain attention and dense feature‑crossing to capture diverse preferences, aggregates click‑tag preferences rather than treating all tags equally, and demonstrates superior offline and online performance over baselines such as YouTube‑DNN, AFM, NFM, DCN, and AUTOINT.
This article details how 58.com leverages deep learning models such as DNN, Wide&Deep, DeepFM, DIN and DIEN, combined with extensive user‑behavior feature engineering, offline vectorization, and online TensorFlow‑Serving pipelines to improve home‑page recommendation click‑through rates and overall platform efficiency.
This article details the design, data pipeline, feature engineering, model development, and iterative optimization of the 58 Tongzhen local feed recommendation system, covering business background, user profiling, recall strategies, ranking models such as XGBoost, XDeepFM, and online learning, and future directions.
This article analyzes why the used‑car search conversion is low, reconstructs user search scenarios from query data, categorizes search intents, identifies pain points across the search funnel, and proposes product‑level redesigns and recommendation strategies to educate vague users and deliver more precise results.
This article describes the design and implementation of an intelligent traffic distribution system for a reading platform, detailing its background, overall architecture, sub-modules such as the small‑traffic experiment platform, near‑line computation, retrieval strategies, pacing algorithms, and how it balances user personalization with content ecosystem growth.
This article introduces Tencent's self‑built Angel distributed machine‑learning platform, describes its architecture and deep‑learning extensions (Parameter Server and AllReduce), explains how it powers the advertising recommendation pipeline with models such as DSSM, VLAD and YOLO, and presents extensive training‑level optimizations that yield multi‑fold performance improvements.
Peng Mo’s talk details Tencent Kankan’s billion‑user feed architecture—layered data, recall, ranking, and exposure control—while addressing real‑time feature generation, massive concurrency, memory‑intensive caching, and fast indexing, and explains solutions such as multi‑level caches, online minute‑level model updates, Redis bloom‑filter exposure filtering, a lock‑free hash‑plus‑linked‑list index, and distributed optimizations that halve latency to under 500 ms and support auto‑scaling and cold‑start handling.
This article reviews the core algorithm modules of recommendation systems from an architectural perspective, discussing offline, near‑line, and online layers, the trade‑offs between personalization, timeliness, and resource consumption, system boundaries, external dependencies, and the practical design of each layer.
The article details the end‑to‑end architecture of Toutiao's massive recommendation platform, covering system overview, content and user feature extraction, model training, recall strategies, evaluation methodology, and content safety mechanisms, while highlighting practical challenges and engineering solutions.
In this interview, Peng Changping, JD's recommendation advertising algorithm lead, explains how large‑scale machine learning and deep learning power JD's e‑commerce recommendation system, detailing its impact on revenue, key technical components, challenges, and future directions.
This article provides a comprehensive technical overview of 今日头条's recommendation system, covering its three-dimensional feature model, algorithm choices, real‑time training pipeline, recall strategies, content analysis, user tagging, evaluation methods, and content‑safety mechanisms.
This article presents a comprehensive overview of Ctrip's hotel recommendation system, covering its technical architecture, data processing pipelines, various ranking and embedding models—including FM, Wide&Deep, DeepFM, and FTRL—deployment methods such as PMML and TensorFlow Serving, offline and online evaluation results, and challenges like cold‑start and diversity.
This article explains how intelligent recommendation differs from generic marketing, introduces association‑rule concepts such as support, confidence, and lift, and provides a step‑by‑step Python implementation using the Apriori algorithm to generate and interpret market‑basket recommendations.
This article provides a detailed introduction to knowledge graphs, covering their definition, why they are needed, the four basic triple types, construction pipelines (including data sources, crowdsourced vs automated methods, and schema versus data layers), storage and query techniques using graph and relational databases, and their practical applications such as enhancing precision, diversity, and explainability in recommendation systems through models like DKN, RippleNet, and graph neural networks.
This article details Likee's short‑video recommendation pipeline, covering the evolution of its feature‑engineering framework, the use of Flink for minute‑level statistical and second‑level session features, the integration of automatic statistical features into DNN models, multimodal feature extraction, and the cache‑tensor technique that dramatically improves online inference performance.
This article details how Jike's recommendation team upgraded from Spark to TensorFlow, introduced a twin‑tower embedding model for recall, deployed it with TensorFlow Serving and Elasticsearch, and achieved a 33.75% lift in user interaction on the dynamic square.
This article presents a comprehensive overview of 58's recruitment recommendation system, detailing business challenges, multi‑stage recall strategies, vector‑based deep retrieval, cost‑sensitive loss design, session optimization, online incremental training, extensive offline and online evaluations, and practical lessons for future improvements.
This article reviews the concept, history, and practical implementations of embedding—including item, image, and user embeddings—and describes various vector‑based recall strategies such as i2i, u2i, clustering, and deep‑learning models used in Tencent's mobile news recommendation platform.
This article shares the author’s experience at Club Factory, describing the business model, growth challenges, macro‑ and micro‑level analysis, and detailed technical breakdowns of recommendation system components—including recall, ranking, user interest modeling, evaluation metrics, and ecosystem considerations—to guide scalable e‑commerce growth.
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.
