This guide explains how to launch the RecBole recommendation system image on the SumW GPU cloud, covering its key features, required setup steps, dependency installation tips, and a one‑line command to run a baseline model on an MLU accelerator.
This article analyzes DeWu's generative recall system, detailing its background, technical design of the Generative and Rerank models, inference workflow, experimental gains in core consumption and diversity metrics, and future engineering directions such as framework migration, LLM integration, and multimodal generation.
X platform has open‑sourced its new "For You" recommendation system, revealing a Grok‑based Transformer architecture, detailed module breakdown, seven‑step content ranking pipeline, and the strategic motivations behind the unprecedented move toward algorithmic transparency and community‑driven improvement.
X (formerly Twitter) has open‑sourced its “For You” recommendation algorithm, revealing a Grok‑based Transformer that merges on‑platform and off‑platform content, removes manual features, and scores posts through a multi‑stage pipeline with candidate sourcing, hydration, filtering, scoring, and selection.
The article dissects X’s newly open‑sourced ‘For You’ feed algorithm, detailing its Rust and Python implementation, the Home Mixer pipeline, candidate sourcing, Grok‑based scoring, and extensive filtering, showing how machine‑learning predicts user interactions and shapes the content you see.
This article summarizes three cutting‑edge AI search and recommendation techniques—Alibaba Cloud's Agentic RAG architecture, Huawei's LLM‑enhanced recommendation system evolution, and Baidu's generative ranking model GRAB—detailing their challenges, design choices, performance gains, and practical deployment insights.
The article outlines a complete open‑source architecture for a short‑video platform, detailing front‑end player features, personalized recommendation, multi‑mode playback, social interactions, monetization models, cross‑platform technology choices, backend micro‑services, CDN, caching, security, and solutions to high‑concurrency and synchronization challenges.
This article details BMW's digital transformation journey using Sensors Data, covering the background of rapid app growth, the cross‑regional data collection challenges, the systematic solution architecture—including mapping, preprocessing, and historical data migration—and the resulting business impact and future AI‑driven roadmap.
This article walks through the design and implementation of a simplified TikTok‑style recommendation system using Spring Cloud microservices, covering business requirements, service decomposition, Eureka registration, Feign clients, Redis and Kafka integration, circuit‑breaker fallback, and end‑to‑end testing.
This article walks through designing and implementing a simplified TikTok recommendation system using Spring Cloud microservices, covering business requirements, service decomposition, project setup, Eureka registration, Kafka and Redis integration, Feign clients, circuit‑breaker fallback, testing, and key deployment considerations.
This article details Xiaohongshu's multi‑stage recommendation pipeline—using massive multi‑modal pre‑training, long‑sequence modeling, real‑time context features, reinforcement learning and online deep learning—to precisely surface valuable content, address cold‑start challenges, and break information bubbles for billions of users.
OneRec, an end-to-end generative recommendation system from Kuaishou, uses an encoder-decoder architecture, reward-based preference alignment, and reinforcement learning to dramatically improve video recommendation efficiency, boosting user engagement and reducing operational costs while achieving scaling-law performance comparable to large language models.
The OneRec system from Kuaishou replaces traditional cascade recommendation pipelines with an encoder‑decoder architecture, leverages reward‑based preference alignment via reinforcement learning, achieves ten‑fold FLOPs gains, cuts operational costs by 90%, and delivers significant user‑engagement improvements across short‑video and local‑service scenarios.
This article presents a novel minute‑level future‑reward framework with dual‑delay incentives, activity‑based attribution, multi‑task delayed modeling, and sequential streaming training that dramatically improves user retention prediction accuracy and real‑time performance in large‑scale recommendation platforms.
This article explains JD's complex recommendation system data pipeline—from indexing, sampling, and feature engineering to explainability and real‑time metrics—highlighting challenges such as data consistency, latency, and the use of Flink for massive, low‑latency processing.
To meet the ultra‑low latency demands of modern recommendation systems, the authors designed a distributed batch‑query architecture featuring the NeighborHash optimization—a cache‑line‑aware hash table that reduces memory accesses, combined with NVMe‑backed storage and AMAC techniques, achieving high throughput and near‑optimal bandwidth utilization.
The article details how edge intelligence is applied to the Vivo official app to improve product recommendation on the smart‑hardware floor by abstracting the problem, designing feature engineering pipelines, training TensorFlow models, converting them to TFLite, and deploying inference on mobile devices, while also covering monitoring and performance considerations.
This article describes how Huolala created a data‑driven, automated testing platform to evaluate and improve loading and unloading recommendation points, covering background challenges, a multi‑layer evaluation framework, offline and online testing methods, metric design, result analysis, smart alerting, and future CI/CD integration.
The article presents a comprehensive adaptive degradation and automatic recovery framework for JD Alliance's recommendation system, designed to handle high‑frequency, instantaneous traffic surges during large promotions by combining real‑time monitoring, Wilson‑interval‑based timeout correction, scenario‑aware control, traffic slicing, linear‑programming‑driven chain optimization, and low‑cost business‑agnostic APIs, achieving over 90% reduction in traffic loss and zero incidents.
This article examines the challenges of handling high‑frequency, instantaneous traffic spikes in JD Alliance's recommendation system during major sales events and presents an adaptive, automated degradation and recovery framework that minimizes recommendation loss while maintaining system stability.
蝉妈妈 migrated its recommendation platform to Alibaba Cloud Serverless Spark and Milvus, replacing traditional vector search and Spark clusters, achieving 40% faster offline tasks, 80% lower failure rates, significant cost savings, and scalable, low‑latency similar‑product retrieval for personalized marketing.
This article explores how applying mathematical modeling and fuzzy evaluation to Chinese language proficiency tests like PEYEL can create personalized learning pathways, improve feedback loops, and bridge the gap between assessment results and actionable teaching strategies.
In 2024 DGraph upgraded its architecture by splitting single clusters into multiple business‑specific clusters, adopting a sharded active‑active topology, and replacing its 1:N thread‑pool with an M:N grouped execution model that uses atomic scheduling, while parallelizing FlatBuffer encoding, streamlining SDK conversions, adding DAG debugging, timeline analysis, and dynamic sub‑graph templates to boost scalability, stability and developer productivity.
The article explains how Douyin's recommendation system uses machine‑learning and deep‑learning models to predict user actions, assign value weights, and dynamically adjust scores, highlighting both its efficiency in large‑scale content distribution and its inherent limitations compared to human understanding.
This article walks through business, conceptual, logical, and physical modeling steps to design a recommendation system architecture, detailing value propositions, workflow decomposition, component breakdown, and technology choices to meet reliability, low‑latency, and scalability requirements.
The article examines Trump’s executive order on TikTok, the platform’s demand to sell half its equity to a U.S. entity, and delves into the sophisticated AI‑driven recommendation algorithms—highlighting the Monolith real‑time system, online training, and research that explain TikTok’s addictive success.
NeighborHash is a distributed batch‑query architecture for real‑time recommendation systems that combines a cache‑line‑optimized hash table—featuring Lodger Relocation, bidirectional cache‑aware probing, and inline‑chaining—with an NVMe‑backed key‑value service, versioned updates, and asynchronous memory‑access chaining to achieve sub‑microsecond, high‑throughput top‑N retrieval.
The article outlines Bilibili’s revamped recommendation data‑flow architecture that eliminates timing and calculation inconsistencies by snapshotting online features, unifying feature computation in a single C++ library accessed via JNI, and orchestrating label‑join and sample extraction through near‑line Kafka/Flink pipelines, with further performance gains and Iceberg‑based future extensions.
This article reviews the evolution of large‑model recommendation techniques, analyzes the specific challenges of health‑oriented e‑commerce recommendation, and details practical deployments such as LLM‑enhanced cold‑start recall, DeepI2I expansion, and scaling‑law‑driven CTR models within JD Health.
This article describes the definition, architecture, and practical applications of explainability in JD's retail recommendation system, covering ranking, model, and traffic explainability, system challenges, data infrastructure, and specific techniques such as SHAP and Integrated Gradients for interpreting model decisions.
This article reviews the role of pre‑ranking in multi‑stage recommendation pipelines, compares dual‑tower and fully‑connected DNN models, discusses negative and positive sample selection strategies, and presents Zhuanzhuan's practical improvements in model architecture and traffic‑pool allocation to boost precision and diversity.
This guide walks through five real‑world Elasticsearch use cases—including full‑text product search with highlighting, centralized log collection and analysis, personalized video recommendation, price‑range aggregation for e‑commerce, and geo‑location restaurant search—detailing index design, query syntax, Docker setup, and front‑end integration.
Bilibili’s game live‑stream platform implements interactive features ‘Play Together’ and ‘Help Me Play’ by using Redis ZSET queues, MySQL persistence, real‑time streamer recommendation, ticket‑based purchase flows, state‑machine order handling, and comprehensive monitoring to ensure reliable, scalable viewer‑streamer gameplay collaboration.
This article details how NetEase Cloud Music leverages multimodal content understanding—using audio models like MusicCLIP and Audio MAE and image‑text fusion via FLAVA—to improve recommendation performance for new content and new users, covering system architecture, cold‑start solutions, and future AI‑driven directions.
This article examines Baidu's AI‑driven overhaul of its app personal center, detailing the problems of the legacy design, the innovative container and recommendation framework, card‑based content structures, dialogue integration, experimental outcomes, and future design insights for AI‑enhanced user experiences.
EasyRec, Alibaba Cloud’s modular recommendation framework, unifies configurable data, embedding, dense, and output layers on MaxCompute, EMR, and DLC, and speeds training with deduplication, EmbeddingParallel sharding, lock‑free hash tables, GPU embeddings, and AMX BF16, while inference benefits from operator fusion, low‑precision AVX/AMX kernels, compact caches, batch merging, and network compression, enabling real‑time online learning and delivering higher recommendation quality at lower cost in e‑commerce.
This article describes how JD Alliance built an adaptive degradation and automatic recovery framework for its recommendation system to handle sudden, large‑scale traffic spikes during major sales events, ensuring stability while minimizing recommendation loss through real‑time monitoring, scenario‑aware control, and linear‑programming‑based pipeline orchestration.
This article presents NetEase Cloud Music's multi‑scenario recommendation modeling work, detailing background, overall system architecture, key modules, modeling goals, technical difficulties, performance improvements, future outlook, and a comprehensive Q&A session that addresses practical deployment challenges.
This paper identifies three challenges of applying large language models to recommendation systems and proposes an Item Language Model that combines an item encoder with a frozen LLM, demonstrating through extensive experiments that language‑item alignment and interaction knowledge significantly improve conversational recommendation performance.
The article introduces an AI‑powered “Volunteer Assistant” built on the agentUniverse multi‑agent framework, detailing how it outperforms existing tools by integrating a specialized SOP, multi‑agent collaboration, and employment‑market analysis to provide precise, personalized college‑major recommendations for high‑school graduates.
This article details the evolution of Tencent News from a portal‑style content display to a personalized recommendation engine, describing the legacy architecture problems, the design goals, the new modular and scalable architecture, feature platform improvements, debugging tools, and stability measures that together increased availability to 99.99% and cut costs by over 60%.
Tencent News upgraded its recommendation architecture by consolidating data platforms, redesigning index and feature services, adopting DDD and Lambda/Kappa patterns, and adding robust debugging and stability measures, which boosted availability to 99.99%, cut CPU, memory and cost by over 60%, and accelerated development and experiment cycles.
This article explains how to design and implement a recommendation system by leveraging a four‑layer cloud‑native stack, covering virtualization, micro‑service migration, service governance, elasticity, cloud‑native business capabilities, and chaos‑engineering‑based stability practices to achieve cost‑effective, high‑performance, and reliable recommendation services.
This article presents Baidu's comprehensive approach to cold‑starting new content in its large‑scale feed recommendation system, covering the definition and challenges of content cold start, algorithmic practices, ID feature optimizations, traffic control mechanisms, experimental design, and key Q&A insights.
This article details the end‑to‑end deep‑learning pipeline used by 58.com for real‑estate recommendation, covering business background, a six‑layer architecture, vector‑based recall, various embedding and ranking models, multi‑task and multi‑scenario optimization techniques, and future directions for large‑model integration.
This article explains how Tubi monitors the quality of its recommendation system by identifying potential failure points, tracking key data streams such as model input, final recommendation output, and training data, and designing a scalable, real‑time monitoring solution with clear protocols and extensible metrics.
This article explores how Laravel can serve as a flexible backend platform for integrating artificial intelligence and machine learning technologies—such as predictive analytics, chatbots, image/video analysis, and recommendation systems—by presenting practical code examples, discussing opportunities, challenges, and best‑practice tools.
This article presents a comprehensive overview of how to design and implement a recommendation system using cloud‑native technologies, covering the cloud‑native stack, system architecture, key design considerations such as virtualization, micro‑service migration, service governance, resilience, and stability through chaos engineering.
This article explains how Kuaishou tackled the growing complexity of its recommendation system by developing the Dragonfly strategy engine framework, detailing the challenges, architectural abstractions, DSL-based workflow composition, data handling, ecosystem tools, and future development plans.
This article explains the role of A/B experiments in recommendation systems, outlines their workflow, shares practical tips and parameter design strategies, and demonstrates how to use experiment parameters and feature flags for efficient testing, optimization, and full‑scale deployment.
Xiaohongshu’s recommendation system uses large‑scale multimodal embeddings, dual‑tower and graph models, and diversity techniques like DPP and SSD to quickly surface high‑quality user‑generated content, enabling ordinary users to gain visibility while balancing personalization, exploration, and efficient LLM‑augmented pipelines.
This article explains the fundamentals of graph neural networks and graph representation learning, outlines how graphs enhance recommendation systems, and details OPPO's practical implementation of a hybrid dual‑tower and graph sub‑network model to improve recall and ranking performance.
This article introduces the use of graph models in Baidu's recommendation system, covering graph fundamentals, common graph algorithms such as graph embedding and graph neural networks, the evolution of the Feed graph model, and its subsequent promotion across multiple product lines.
The article explains how the data team adopted the open‑source machine‑learning platform MindsDB to simplify data integration, enable SQL‑based model training and inference, manage model versions, and dramatically shorten recommendation system development cycles, achieving up to 30% efficiency gains.
This article presents Alibaba's enterprise knowledge recommendation system, detailing its role in digital transformation, the challenges of long‑document recommendation, the multi‑layer architecture spanning feature, engine, ranking, and functional layers, various recall strategies, progressive ranking models, and the integration and evaluation of large language models for improved recommendation performance.
OPPO revamped its advertising recall system by replacing a latency‑prone directional pipeline with an ANN‑based full‑ad personalized architecture, employing a dual‑tower LTR model, multi‑path auxiliary branches, refined offline metrics, price‑sensitive and hard‑negative sampling, and hybrid joint training, which together boosted ARPU by about 15%.
This article presents OPPO's advertising recall system, detailing the transition from the legacy architecture to a new ANN‑based design, model selection criteria, offline evaluation metrics, sample optimization techniques, and various model improvements that together achieved significant ARPU gains.
This article explores the end‑to‑end design of a modern dating platform, covering requirement analysis, micro‑service architecture, gateway routing, sharded MySQL, CDN caching, matchmaking, recommendation scoring, high‑concurrency strategies, load balancing, database optimization, message queues, and spatial proximity algorithms such as grid, quadtree, and GeoHash.
This article describes how Kuaishou built the Dragonfly strategy engine framework—covering problem analysis, architecture design, DSL-based workflow orchestration, process and data abstractions, ecosystem tools, and future plans—to solve the scalability, coupling, and maintenance issues of its rapidly expanding recommendation services.
This article presents NetEase Cloud Music's multimodal cold‑start solution, detailing the problem background, feature selection using CLIP, two modeling approaches (I2I2U indirect and U2I DSSM direct), contrastive learning enhancements, interest‑boundary modeling, and evaluation results showing significant gains in user engagement.
The article describes how NetEase News' recommendation team replaced a slow, batch‑oriented data‑warehouse pipeline with a Flink‑based, Apache Paimon real‑time wide‑table solution that supports partial updates, reduces latency from hours to minutes, and lowers processing costs while handling both deduplication and non‑deduplication recommendation scenarios.
This article presents an in‑depth overview of Tencent's TRS automatic hyperparameter tuning, covering background, challenges, the evolution from Bayesian optimization to evolution strategies and reinforcement learning, a systematic platform solution, real‑world deployment results, and a Q&A session.
The article details how Autohome tackled the cold‑start problem for newly generated content by redesigning the recommendation pipeline, introducing multi‑path recall, refining ranking and re‑ranking formulas, and applying strategic controls, resulting in a rise of cold‑start success rate from 27% to over 99% and a CTR increase from 5% to 14%.
This article describes the design, construction pipeline, and practical applications of a financial event knowledge graph, covering background challenges, multi‑layer modeling, information‑extraction techniques, and use cases such as institutional risk monitoring, wealth‑management recommendation, and industry‑chain analysis.
This article details how AutoHome's homepage pop‑up leverages precise targeting, extensive feature engineering, and multi‑stage DeepFM‑based models with attention and LHUC modules to accurately identify car‑buying users, improve vehicle‑series recommendations, and achieve a 355% conversion rate increase.
This article presents Xiaohongshu's evolving recommendation engineering architecture, detailing its modular backend design, graph-based Ark framework, hot deployment mechanisms, and the challenges and solutions for scaling personalized content delivery in a fast‑growing mobile platform.
This article provides a concise overview of recommendation systems, covering their definition, core framework of recall, ranking, and re‑ranking, various recall strategies including multi‑path and vector‑based methods, similarity calculations, and practical implementation details such as AB testing and code examples.
This paper presents TWIN, a two-stage interest network that aligns the similarity metrics of coarse‑grained and fine‑grained modules to improve lifelong user behavior modeling for CTR prediction in large‑scale online recommendation systems.
Airbnb upgraded its neural‑network ranking system by adding a similarity network that penalizes duplicate‑like listings, enabling the algorithm to present a more diverse set of options, which boosted booking rates, value, and five‑star ratings, demonstrating that reduced result similarity improves overall search quality.
This article details the design, componentization, platformization, and low‑code tools of a recommendation‑algorithm PaaS that streamlines development, supports diverse business scenarios, and accelerates delivery of personalized recommendation capabilities across multiple product lines.
The Bilibili Game Center recommendation system combines a unified feature platform, multi‑stage recall, ranking and re‑ranking models, online services, and AB experimentation to deliver personalized game suggestions, resulting in up to 78% higher click‑through, 76% higher conversion, and substantial increases in user engagement and revenue.
Meituan’s food‑delivery team built a novelty‑focused recommendation pipeline—combining dual‑tower recall, novelty‑aware ranking, personalized mixed‑ranking weights, and reinforcement‑learning insertion—to surface merchants unseen by users, achieving 19% higher exposure novelty, 25% more order novelty, and improved ratings while keeping RPM loss under 0.5%.
This article explains the design, implementation, and iterative optimization of Boss Circle's recommendation engine, covering the initial simple ranking, the introduction of Elasticsearch‑based scoring, multi‑pool data sources, machine‑learning experiments, real‑time feature handling, and future personalization challenges.
Twitter’s recently open‑sourced recommendation system is dissected, covering its overall architecture, graph‑based data and feature engineering, recall pipelines (in‑network and out‑of‑network), coarse and fine ranking models, mixing and re‑ranking stages, as well as the supporting infrastructure and code examples.
This paper presents RLUR, a reinforcement‑learning algorithm that models user‑retention optimization as an infinite‑horizon request‑based Markov Decision Process, addressing uncertainty, bias, and delayed reward challenges to directly improve retention, DAU, and engagement in short‑video recommendation platforms.
Taobao’s Daily Good Store tackles a two‑level recommendation challenge by jointly ranking shops and their items through a dual‑link system enhanced with a novel scatter‑score metric, personalized category scattering via Earth Mover’s Distance, beam‑search optimization, and UI upgrades, delivering higher efficiency, relevance, diversity, and ecosystem health.
This article examines the open‑source Twitter recommendation algorithm released by Elon Musk, detailing its main services, machine‑learning models, data sources, programming languages, and the GitHub repositories that host the core components such as SimClusters, TwHIN, rankers, and the Rust‑based navi framework.
Xiaohongshu transformed its recommendation pipeline from daily to minute‑level updates by redesigning recall, ranking and feature‑joining components, deploying a base‑plus‑incremental training scheme, migrating Spark to Flink, rewriting services in C++, and optimizing RocksDB, which yielded over 10% longer dwell time, 15% more interactions and roughly 50% higher new‑note efficiency.
This article details how Ximalaya’s AI Cloud adopted the open‑source HybridBackend framework to overcome sparse data access and distributed training bottlenecks, achieving multi‑GPU utilization gains, faster model training, and significant cost reductions across its recommendation services.
The article describes WeChat Games’ scalable recommendation system, detailing its four‑component architecture—offline ML platform, unified management, online DAG‑based engine, and peripheral services—along with a hybrid algorithm library, feature engineering, real‑time monitoring, and solutions that boost engagement across diverse game recommendation scenarios.
The article describes a modular traffic‑control platform for e‑commerce search and recommendation that lets operators quickly adjust strategies for emergencies, cold‑start items, and experiments, replacing costly multi‑team development with a unified operation center, service center, data hub, algorithmic PID controller, real‑time metrics, independent recall chain, and cross‑scene AB testing, while outlining future extensions.
KGraph, developed by Kuaishou since late 2019, is a self‑built graph platform that supports massive social, e‑commerce, and security workloads, offering a distributed KV storage, high‑performance RPC framework, and advanced graph modeling to achieve tens of millions of QPS and low latency for real‑time recommendation and offline graph analytics.
This article presents a comprehensive overview of Kuaishou's large‑scale recommendation system, detailing its pipeline, unique characteristics, CTR model improvements, the PPNet personalization network, multi‑domain multi‑task framework, short‑ and long‑term behavior sequence modeling, and the challenges of handling billions of features and trillions of parameters.
The article outlines a comprehensive quality‑assurance framework for algorithm models on the Yanxuan platform, detailing lifecycle stages, common issues, and a unified platform that automates bad‑case mining, model‑effect monitoring, latency tracking, and pipeline validation to ensure reliable deployment across search, recommendation, marketing, bidding, and forecasting applications.
The article provides a comprehensive technical overview of Toutiao's growth, data collection, user modeling, recommendation engine, storage solutions, message push system, and its micro‑service and virtualized PaaS architecture, highlighting the massive scale and engineering practices behind the platform.
This interview compiles expert opinions on the end‑to‑end recommendation system pipeline—including architecture, data collection, user profiling, content structuring, feature engineering, recall strategies, ranking algorithms, multi‑objective optimization, multi‑modal fusion, re‑ranking, cold‑start solutions, evaluation metrics and real‑world applications—highlighting the technical challenges and practical solutions.
The Xianyu recommendation system, built by backend expert Wan Xiaoyong, evolved from offline scoring to a full‑graph, serverless recall‑ranking pipeline that tackles C2C uncertainties through centralized feature engineering, model compression, staged deployment, flexible experimentation, robust governance, and plans for automated attribution and interpretability.
The article details Meituan's deployment of large-scale heterogeneous graph recall for in‑store recommendation ads, covering full‑scene graph construction, graph pruning, dynamic negative sampling, spatiotemporal sub‑graph fusion, and performance optimizations that together raise offline hit‑rate by over 5% and online revenue per search by 10‑15%.
Xiaohongshu’s in‑house LarC platform powers real‑time, multimodal recommendation, life‑search, and generative‑AI commercial content for its 200 million‑user community by processing billions of daily feedback samples, employing conflict‑free parameter servers, diversified sequence modeling, and large‑scale representation learning to deliver personalized, fresh, and diverse user experiences.
This article details the design and optimization of Jimu’s card‑based stranger‑social recommendation system, covering product background, gameplay flow, technical challenges in strategy and engineering, a multi‑task AI ranking model, vector recall improvements, and the resulting performance gains.
The article presents a generalized recommendation platform for Xianyu Marketplace that consolidates feature processing, candidate generation, recall, scoring, and experiment management into shared core components, enabling rapid onboarding of new scenarios, reducing engineering effort, and delivering over 8% CTR lift and 10% more impressions.
This article presents how Ant Group leverages graph machine learning, including knowledge graph, social network, and cross-domain techniques, to enhance recommendation for low-activity users across scenarios such as fund, coupon, and waistband recommendations, detailing model architecture, challenges, solutions, and experimental results.
This article introduces the fundamentals of recommendation systems, explains Zhuanzhuan's main recommendation scenarios and pipeline, and details three graph embedding methods—DeepWalk, node2vec, and EGES—along with their practical implementations in recall and coarse‑ranking stages.
This article explains Hulu's server‑side feature logging system that aligns online and offline recommendation features, measures and mitigates feature drift caused by data source, timing, and code differences, and improves model performance while reducing resource consumption.
This article explains the fundamentals of recommendation systems, introduces collaborative filtering (both user‑based and item‑based), derives the matrix‑factorization model with ALS optimization, provides a complete Python implementation, and demonstrates how to apply Spark ALS in both demo and production environments.
The paper describes NetEase Cloud Music’s LOOK live‑streaming recommendation system for the song‑playback page, which combines millisecond‑level real‑time feature pipelines, multi‑target optimization (click, watch, gift, comment) via ESMM+FM and MMoE models, GradNorm‑based loss fusion, and a multimodal avatar‑text‑host ranking model, achieving double‑digit CTR and CTCVR gains while balancing producer and consumer retention.
The article details a step‑by‑step evolution of an e‑commerce platform’s top‑slot recommendation system, moving from a DCN‑mix single‑objective model through BST‑based dynamic features, position‑bias debiasing, multi‑task MMoE learning, and finally BST with target‑attention, each yielding measurable CTR, conversion, and user‑value gains.
This article details Kuaishou's practical advances in short‑video recommendation, covering the CTR PPNet model, multi‑domain multi‑task learning, short‑ and long‑term user behavior sequence modeling, and the resulting performance gains in user engagement and app usage.
In a July 2022 keynote, Halu’s senior algorithm expert Wang Fan outlined the construction of an intelligent transaction system for its car‑pool service, detailing business challenges, a decomposition into matching, pricing, marketing and arbitration, a recommendation‑pipeline architecture, and three‑stage algorithm evolution that boosted order volume by over 20 %.
This article presents JD's comprehensive technical solution for boosting recommendation ad ranking by addressing cold‑start, shallow user interest extraction, and insufficient global data through a variational feature learning framework, enhanced user‑interest networks, and full‑domain collaborative modeling, achieving over 1% AUC gain and notable revenue growth.
