The May 15 open‑source release of X’s For‑You recommendation system reveals a full pipeline—from query hydration and candidate sourcing to multi‑stage scoring—showing that the platform predicts a range of user actions, emphasizes content‑level signals, and offers creators concrete guidance to improve visibility.
This article dissects the end‑to‑end architecture of a billion‑scale short‑video platform, detailing layered design, core services such as upload, transcoding, recommendation, interaction, storage, and the key challenges of massive video storage, high‑concurrency streaming, low‑latency playback, and real‑time recommendation reliability.
HyFormer, a novel hybrid Transformer framework introduced by ByteDance’s TikTok search team, integrates sequence modeling and feature interaction into a unified alternating optimization process, enhancing representation power and scaling efficiency for ultra‑long user behavior sequences and high‑dimensional heterogeneous features, leading to significant offline and online performance gains.
OxygenREC presents a fast‑slow thinking, instruction‑following generative framework that overcomes latency, reasoning, and multi‑scene scalability challenges in e‑commerce recommendation, delivering unified training, low‑latency inference, and significant business impact across JD.com scenarios.
The article examines JD’s AI‑powered shopping app, detailing its chatbot‑style interface, intent‑driven workflow, AI‑enhanced product recommendation, multi‑scenario integration such as travel and dining, and the underlying research on Fast‑Slow thinking and the SA‑GCPO algorithm that powers the experience.
This article curates eight AAAI 2026 papers authored by the Meituan research team, covering verifiable stepwise rewards for LLM reasoning, annealing strategies in large‑scale training, process reward models, competence‑difficulty sampling, high‑fidelity visual text rendering, counterfactual fusion, compress‑then‑rank reranking, and cross‑modal quantization for generative recommendation, with direct PDF links for each work.
Kuaishou secured 12 papers at AAAI 2026, covering advances in search and recommendation systems, multi‑camera video generation, multimodal understanding, generative model fundamentals, video large language models, experimental design, and LLM latent‑space reasoning, with three papers highlighted as oral presentations.
MUSE introduces a multimodal search‑based framework that reorganizes tens of thousands of dormant user actions into a unified visual‑semantic interest graph, enabling CTR models to leverage ultra‑long behavior sequences with a 12.6% lift in online performance.
This article reviews three advanced AI search solutions—Alibaba Cloud's Agentic RAG architecture for multi‑modal retrieval, Huawei's LLM‑enhanced recommendation system with factorized prompting, and Baidu's generative ranking model GRAB—detailing their technical challenges, design choices, performance gains, and deployment insights.
Gaode’s G‑Action framework combines large‑language‑model pre‑training with fine‑tuned generative recommendation to predict a user’s immediate action and destination, transforming static map services into a dynamic, context‑aware experience and delivering measurable gains in click‑through and engagement metrics.
This article explains the EasyRec recommendation system’s training and inference architecture, detailing optimization techniques such as embedding parallelism, CPU/GPU placement, XLA and TRT fusion, online learning pipelines, network compression, and real‑world deployment results that dramatically improve throughput and latency.
This article reviews Kuaishou’s two‑year exploration of large‑model techniques in advertising, detailing the content‑domain estimation challenges, how multimodal and LLM approaches improve full‑domain behavior utilization and external knowledge integration, and introducing the COPE product‑content representation framework and the LEARN LLM knowledge‑transfer system.
This article explains why a robust tag system is essential for recommendation and mining strategies, outlines the hierarchy of entity, concept, and theme tags, and provides practical principles, architecture, and step‑by‑step methods for constructing and managing tags in large‑scale data platforms.
This article explains how recommendation systems work, distinguishes online services from offline computation, outlines a typical recommendation flow, and presents a three‑stage evolution (1.0, 2.0, 3.0) with design principles for stability, observability, and efficiency, culminating in a DAG‑based orchestration and traceable execution.
This article explains the EasyRec recommendation system's training and inference architecture, details a series of optimizations for both CPU and GPU pipelines, and describes the online learning workflow that enables real‑time model updates across large‑scale e‑commerce scenarios.
The article presents the Universal Recommendation Model (URM), a large‑language‑model‑based recall framework that integrates world knowledge and e‑commerce expertise through knowledge injection and prompt‑driven alignment, achieving significant offline recall gains and a 3.1% increase in ad consumption while meeting high‑QPS, low‑latency production constraints.
JD’s advertising platform replaces rule‑based recall with a generative large‑model pipeline that unifies e‑commerce knowledge, multimodal user intent, and semantic IDs across recall, coarse‑ranking, fine‑ranking and creative optimization, while meeting sub‑100 ms latency and sub‑¥1‑per‑million‑token cost through quantization, parallelism, caching, and joint generative‑discriminative inference, delivering double‑digit performance gains and paving the way for domain‑specific foundation models.
The article presents a series of technical breakthroughs by JD's advertising team that improve the quality and coverage of AI‑generated ad images through a trustworthy multimodal feedback network, introduce a large human‑annotated image dataset, and enhance creative ranking with offline multimodal representations and online architecture optimizations, ultimately achieving more precise and scalable ad personalization.
This article explains the DPP platform’s overall architecture, its key features for rapid iteration, and the three‑stage evolution of its orchestration engine—from the fixed DPP‑Engine to the flexible BizEngine and finally the graph‑based DagEngine—detailing design decisions, protocols, challenges, and future directions.
Recommendation algorithms, rooted in machine learning and deep learning, transform massive user‑generated data into mathematical models that filter and personalize content, covering traditional collaborative filtering, matrix factorization, cosine similarity, and modern deep models such as Wide & Deep and Two‑Tower retrieval, illustrating their evolution and practical applications.
This article presents a Kuaishou team's non‑autoregressive generative approach for recommendation reranking, detailing its architecture, loss design, experimental validation on Avito and Kuaishou datasets, and online A/B results that earned acceptance at KDD 2024.
This article presents a comprehensive overview of EasyRec’s recommendation system architecture, detailing training and inference optimizations, embedding parallelism, CPU/GPU placement strategies, online learning pipelines, and network compression techniques that together improve scalability, latency, and cost efficiency.
This article highlights eight Meituan research papers accepted at CIKM 2024—spanning self‑supervised sequential recommendation, rating‑consistent explanation generation, CTR prediction via recommendation pre‑training, cross‑domain interest transfer, multimodal vector retrieval, design‑aware poster layout, order‑fulfillment cycle‑time forecasting, and delivery‑scope substitution—offering insights from both internal and university collaborations.
This article presents a comprehensive solution for heterogeneous long‑behavior sequence modeling in advertising recommendation, introducing the TIN backbone, Disentangled Side‑Info TIN, Stacked TIN, and Target‑aware SASRec, along with platform‑level optimizations that enable million‑scale sequences while delivering significant online performance gains.
The article examines how large language models are reshaping JD Retail's e‑commerce search and recommendation pipelines, detailing industry evolution, technical challenges such as knowledge hallucination, intent understanding, personalization, cost, and safety, and presenting JD's end‑to‑end AIGC architecture, data preprocessing, alignment, evaluation, and next‑generation AI search solutions.
This article provides a comprehensive analysis of the HLLM paper, detailing its hierarchical LLM architecture for item and user modeling, the training objectives, fusion strategies, extensive offline and online experiments, scaling behavior, ablation studies, and practical deployment insights in large‑scale recommendation systems.
This article presents JD.com’s multimodal recommendation system architecture, covering content understanding, multimodal ranking and recall models, practical deployment pipelines, and future research directions such as large‑model integration and supply‑side generation, all illustrated with detailed diagrams and Q&A.
This article presents a comprehensive overview of EasyRec's recommendation system architecture, detailing training and inference optimizations, distributed deployment strategies, operator fusion techniques, online learning pipelines, and network-level improvements to enhance performance and scalability.
This article analyzes how Xiaohongshu addresses the decentralization challenge in its recommendation system by strengthening side‑information usage, integrating multimodal signals across the full pipeline, and implementing interest exploration and protection mechanisms, while also outlining future research directions such as generative recommendation and large‑model‑driven user profiling.
This article presents Xiaohongshu's technical solutions for decentralized content distribution, covering the definition of the problem, fast and accurate learning, side‑information modeling, graph‑based multi‑modal fusion, interest exploration and protection, and future research directions such as generative recommendation and large‑model driven user profiling.
This article summarizes Tencent Advertising's recent research on recommendation models, covering comprehensive feature encoding techniques, solutions to embedding dimensional collapse through multi‑embedding paradigms, and novel methods such as STEM and AME to disentangle conflicting user interests across multiple tasks.
Alibaba Auction improves large‑asset recommendation by deploying the multi‑interest vector recall model MIND and the two‑hop PDN model, adapting features and time weighting for unique, high‑value items, using hard‑negative sampling and combined rule‑based and vector similarity, which boosts conversion metrics while revealing filter‑bubble concerns.
This article introduces a Python‑Flask‑MySQL student training plan management system, detailing its features such as visualized progress, SVD‑based course recommendation, forum discussion, simulated enrollment, project structure, environment setup, step‑by‑step installation, configuration, and usage instructions.
This article details eBay's practical experience integrating multimodal data and graph neural networks into its recommendation pipeline, covering pain‑point analysis, a twin‑tower multimodal embedding model with triplet loss and TransH, engineering design, experimental results, and key takeaways for future AI‑driven product development.
This article presents a comprehensive overview of building a user profile tag system—including tag taxonomy, platform architecture, construction methods, update cycles, access patterns, common algorithmic tags, and real‑world applications such as marketing, metric attribution, and A/B testing—illustrated with examples and a detailed Q&A session from a data‑mining senior manager at Qunar.
The Mixed‑Modal Expert Model combines ID features with image and text embeddings through optimized representations and conditional output fusion, dramatically improving advertising recall—especially for long‑tail items—and delivering measurable gains in click‑recall, revenue, CTR, and page views in large‑scale online tests.
This article summarizes the UniSAR framework, which models four types of fine‑grained user transitions between search and recommendation, demonstrates its effectiveness on public datasets, and shows how joint learning improves both search relevance and recommendation quality.
The paper introduces FIN (Fragment and Integrate Network), a novel spatial‑temporal model that extracts multiple sub‑sequences from ultra‑long user behavior logs, applies simplified and multi‑head attention, and fuses them with physically meaningful set operations, achieving up to 5.7% CTR lift and 7.3% RPM improvement in real‑world food‑delivery advertising.
This article highlights three oral papers accepted at WSDM 2024 that address cross‑domain sequential recommendation, extremely sparse feedback denoising recommendation, and automated label crafting for short‑video recommendation, providing their abstracts, author lists, and links to PDFs and source code.
This article details the design, integration, and operational practices of a comprehensive user and item profiling tag system, covering tag taxonomy, construction methods, update cycles, access strategies, algorithmic implementations, and real‑world applications such as marketing, attribution analysis, and A/B testing.
This article details how a Twitter recommendation engine was refactored over three months to improve stability, introduce scalable tooling, redesign material storage and read‑status services, and ultimately raise availability from under 99% to over 99.9% while cutting latency and resource usage.
This article presents Fliggy Travel's end‑to‑end algorithmic solutions for hotel category planning, introduces the LINet group‑recommendation model that incorporates location and travel intent, and details the PETS two‑stage model for selecting hot‑sale hotels under recall constraints, together with experimental results and practical insights.
This article introduces ten data‑mining applications for marketing—including customer segmentation, purchase forecasting, market‑basket analysis, churn prediction, sentiment analysis, response modeling, recommendation systems, brand reputation, competitive analysis, and public‑opinion monitoring—each illustrated with concise Python code examples.
The team upgraded Taobao’s feed mixing by deploying an independent xhuffle service built on the xrec framework, which unifies ad and natural recommendation objectives, decouples strategy from business logic, and uses a serial integration to keep average latency under 30 ms while improving both natural and ad metrics, with plans to extend mixing to short video, live streams, and broader scenarios.
This article explains the role of a strategy product manager, outlines their three core actions, compares the role with client product and data analyst positions, presents the "Dao" (values) and "Shu" (methods) guiding recommendation strategy, and answers practical Q&A on balancing commercial pressure, AI impact, and content creation versus consumption.
The CCIR 2023 conference in Beijing, sponsored by Kuaishou, hosted a short‑video recommendation algorithm frontier research forum where over 100 experts and students shared the latest AI‑driven recommendation technologies, open datasets, and interdisciplinary challenges in short‑video platforms.
This article describes Kuaishou's approach to cold-start content recommendation, outlining the problems addressed, challenges in modeling sparse new videos, and solutions including graph neural networks, I2U retrieval, TDM hierarchical retrieval, bias correction, and future research directions.
This article introduces the Deep Contextual Interest Network (DCIN), a novel CTR prediction model that jointly models clicked items, surrounding display context, and position bias through three modules—PCAM, FCFM, and IMM—showing significant offline AUC gains and a 1.5% online CTR improvement.
This article presents Huawei's research on applying multi‑task and multi‑scenario learning to advertising recommendation, covering background challenges, algorithm taxonomy, detailed designs of TAML, CausalInt and DFFM models, experimental results, and a Q&A session.
BOMGraph introduces a unified heterogeneous graph neural network that jointly models text, image, and similar‑item search across multiple e‑commerce scenarios, using meta‑path‑guided attention, disentangled scenario‑specific and shared embeddings, and contrastive learning to alleviate sample sparsity, achieving consistent offline and online performance gains.
AntM2C is a publicly released, billion‑sample click‑through‑rate (CTR) dataset covering five distinct Alipay business scenarios, providing both ID and rich multi‑modal (text and image) features to enable comprehensive evaluation of multi‑scenario, cold‑start, and multi‑modal CTR models at industrial scale.
The paper introduces SHARK, a two‑component framework that uses a fast Taylor‑expanded permutation method to prune embedding tables and a frequency‑aware quantization scheme to apply mixed‑precision to embeddings, achieving up to 70% memory reduction and 30% QPS improvement in industrial short‑video and e‑commerce recommendation systems.
This article presents a comprehensive design for Zhuanzhuan's home‑page recommendation pipeline, detailing the system architecture, challenges of traffic efficiency and diversity, and a two‑stage solution that applies Proximal Policy Optimization reinforcement learning in the re‑ranking module and Determinantal Point Process optimization in the coarse‑ranking and traffic‑pool stages, followed by offline simulation, online deployment, and evaluation metrics.
This article presents Kuaishou's approach to solving the content cold-start problem by analyzing its impact on video growth, detailing the challenges of sparse and biased training data, and describing a suite of graph‑neural‑network, I2U/U2I, TDM, and debiasing techniques that improve early video exposure and long‑term ecosystem health.
This article introduces Social4Rec, a video recommendation algorithm that tackles user cold‑start problems by extracting and integrating social interest information through coarse‑ and fine‑grained interest extractors, attention‑based fusion, and extensive offline and online experiments demonstrating significant CTR improvements.
This article presents Xiaohongshu's evolving recommendation system architecture, detailing the challenges of massive user‑generated content, the adoption of a graph‑based Ark framework for modular and scalable business logic, and the implementation of hot‑deployment techniques to accelerate algorithm iteration and reduce downtime.
The paper introduces DCAM, a dynamic context‑adaptation model that automatically selects the most effective spatiotemporal features for online food recommendation, showing that more features or naïve self‑attention do not guarantee gains, and achieving superior offline AUC and online CTR improvements over existing state‑of‑the‑art methods.
The paper introduces PEPNet, a plug‑and‑play network that tackles the domain‑seesaw and task‑seesaw problems in multi‑scenario recommendation by using a gated personalization module (GateNU) together with embedding‑level (EPNet) and parameter‑level (PPNet) personalization, and demonstrates its superiority through extensive offline and online experiments on Kuaishou data.
Eight Alibaba Mama team papers accepted at CIKM 2023 present advances such as task‑specific bottom‑representation networks for recommendation, a unified GNN for multi‑scenario e‑commerce search, multi‑slot bid shading, consistency‑oriented pre‑ranking, bias‑mitigating CTR prediction, efficient progressive‑sampling self‑attention, delayed‑feedback conversion modeling, and hybrid contrastive multi‑scenario ad ranking.
By completely rebuilding its topic system in 2021, Bilibili introduced an AI‑driven pipeline that automatically discovers, creates, ranks, and populates hot and cold‑start topics using real‑time metrics, rule‑based and vector‑based recall, dramatically boosting content relevance, user interaction, and operational efficiency across the platform.
This article presents a comprehensive overview of the push precision recommendation system, detailing its data processing pipeline, machine‑learning‑driven algorithms, modular architecture—including offline, near‑real‑time, and push layers—and subsequent system iterations, optimizations, visual monitoring platforms, and future development directions.
This article introduces user path analysis and the SessionAnalytics open‑source framework, covering business scenarios, data processing techniques, algorithmic mining methods, technical architecture, implementation details, comparisons with event‑based analysis, and a comprehensive Q&A for practical deployment.
This article explains how Xianyu’s product recommendation system tackles massive data, diverse business scenarios, and engineering challenges by designing a unified recommendation middle‑platform that abstracts data, recall, ranking, and re‑ranking stages, enabling rapid scene onboarding and scalable model iteration.
This article presents OPPO's Andes Smart Cloud team's intelligent growth algorithm architecture, covering industry background, data pipelines, model designs such as uplift, PU‑learning, multimodal AIGC, and their practical applications in content supply, recommendation, precise audience targeting, and ad bidding, followed by a summary and Q&A.
Social4Rec introduces a social interest‑enhanced video recommendation framework that tackles user cold‑start by extracting coarse‑ and fine‑grained social interests via a self‑organizing neural network and meta‑path neighborhood aggregation, integrating these embeddings with a YouTube DNN model to improve CTR and AUC.
This article proposes a graph‑based node representation method that combines static attribute graphs and dynamic interaction graphs with multi‑level attention to alleviate user and item cold‑start problems in recommendation systems, achieving notable AUC improvements on sparsified MovieLens datasets.
Qianchuan is a unified recommendation platform that consolidates numerous low‑traffic, diverse scenarios into a five‑layer architecture—service, access, DPP, algorithm, and infrastructure—addressing challenges of varying products, goals, strategies, recommendation types, and limited resources through flexible product selection, multi‑goal support, advanced recall and ranking models, and extensible, low‑cost algorithms, while planning broader scene coverage, bias reduction, and componentized, reproducible solutions.
This article explains how Alipay’s homepage team collaborates with the edge‑intelligence team to use real‑time client‑side behavior data and algorithm platforms, transforming refresh strategies across time, space, and event dimensions, improving recommendation efficiency, reducing duplication, and delivering measurable performance gains.
The paper proposes a dual‑interest decomposition head‑attention model that uses a feedback‑aware encoding layer, a factorized head attention mechanism, and separate positive/negative interest towers to improve sequence recommendation performance on short‑video and e‑commerce datasets.
Daily Good Shop improves e‑commerce recommendations by first ranking products with long‑term user behavior models, assembling top items into cards, then ranking those cards using a shared‑bottom multi‑task network that jointly predicts click, subscription and lead‑IPV, and finally re‑ranking card sequences via beam‑search, yielding over 2 % more clicks, 34 % more subscriptions, 33 % more lead‑IPV and 22 % longer dwell time.
This article describes the architecture, data processing, city‑intent modeling, resource recall strategies, and multi‑task ranking models—including PLE‑CGC and ESMM—used to improve click‑through and conversion rates of the Tongcheng travel app's homepage waterfall‑flow recommendation, and outlines experimental results and future optimization directions.
Xiaohongshu’s end‑to‑end intelligent creative platform extracts high‑quality images, generates diverse titles with RED‑pretrained GPT‑2/T5 models, and selects the best ads using a UCB‑based multi‑armed bandit that balances CTR uplift, revenue and user‑experience, while employing position‑corrected metrics and a scalable dual‑tower DNN to boost long‑tail performance and overall revenue.
This article compares advertising and recommendation algorithms, highlighting distinct optimization goals, model design focuses, training methods, implementation principles, auxiliary strategies, and model characteristics, emphasizing how ads aim to increase revenue while recommendations prioritize user engagement and diversity.
The article introduces FiBiNET, a CTR prediction model that incorporates a SENet module for dynamic feature‑importance learning and a bilinear‑interaction layer for enhanced second‑order feature interactions, then details its improved variant FiBiNET++ which reduces parameters with Bi‑Linear+ and an enhanced SENet+.
On January 9 from 19:00 to 20:00, algorithm engineer Zhang Yuanliang will present Taobao’s scenario-adaptive, self-supervised multi-scenario recommendation model, detailing its architecture, experimental results, and practical deployment for improving personalized item recall across diverse user contexts.
This article presents Xiaomi’s e‑commerce recommendation research, addressing four key challenges—sample selection bias, data sparsity, delayed feedback, and knowledge inconsistency—by introducing the Entire Space Delayed Feedback with Cross‑Task Knowledge Distillation (ESDC) model, which combines causal inference, cross‑task distillation, twin networks, and uncertainty weighting to improve CVR prediction and achieve a 15% GMV lift over the baseline.
The Adaptive Domain Interest Network (ADIN) introduces a shared backbone with scenario‑specific subnetworks, domain‑specific batch normalization and SE‑Block attention to capture both commonalities and divergences across recommendation scenarios, and, combined with self‑supervised training, consistently outperforms baselines, delivering a 1.8% revenue lift in Alibaba’s display‑ad platform and now runs in production.
The article details the architecture and techniques of ZhiZhuan's low-result search module, explaining how it combines ElasticSearch hard matching and sBert semantic vector soft matching, along with sophisticated negative sample strategies, to improve recommendation coverage and user experience.
The article examines repurchase strategies for game item recommendations, analyzing various recommendation scenarios, their specific challenges, item classification based on purchase density and repurchase rates, and practical guidelines for applying the strategy across permanent shop, limited‑time gift packs, and refreshable recommendations.
The article outlines how search and recommendation systems have evolved from simple hot‑list displays to sophisticated, data‑driven pipelines comprising recall, fine‑ranking and re‑ranking stages, describes an integrated low‑code engine with standardized features, configurable components and intelligent modules that enable rapid deployment across many scenarios, delivering notable CTR, GMV and engagement gains at 哈啰.
At KDD 2022, Tencent Advertising introduced the Multi‑Virtual‑Kernel Experts (MVKE) model, a novel multi‑task architecture that replaces traditional dual‑tower structures with virtual‑kernel experts and a gating mechanism to jointly learn user preferences across multiple actions and topics, achieving superior offline and online performance.
This article presents a comprehensive overview of building an automotive domain knowledge graph—from ontology design, data acquisition, and graph schema construction using JanusGraph, to its practical use in cold‑start, explanation, and ranking stages of recommendation systems—highlighting challenges, solutions, and performance benefits.
Alibaba Mama’s technical team presented ten CIKM‑2022 papers that introduce novel advertising and recommendation methods—including adaptive domain networks, neural‑metric ANN search, control‑based livestream bidding, graph‑based relevance learning, hierarchical ad exposure, knowledge‑extraction pretraining, traffic forecasting, overfitting analysis, adaptive sparsity, and visual debiasing—each deployed to boost revenue and performance on Alibaba’s platforms.
This article presents JD.com's end‑to‑end intelligent creative system, covering the background of content‑driven e‑commerce, a multi‑stage content understanding pipeline, AI‑powered video, image and copy generation, multimodal creative selection and distribution, and real‑world business impact.
This article presents Hello's Intelligent Creative project, detailing its business motivations, system architecture, algorithmic choices such as seq2seq, VAE, GAN, and pre‑trained models, the implementation of material libraries, tagging, recall strategies, a creative racing model, performance gains, and future challenges.
Cold‑Transformer tackles the user cold‑start problem by introducing an Embedding Adaption layer that refines sparse user embeddings using context‑aware fused behavior sequences and a label‑encoding scheme, preserving a dual‑tower design and achieving state‑of‑the‑art performance on public and industrial datasets.
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 article presents a request‑granularity quality model for Meituan's integrated recommendation system, linking data tables, algorithm models, services, and user requests, and details its metrics, defect taxonomy, calculation formulas, data‑lineage expansion, implementation, alert routing, and operational outcomes.
This article details the design, data pipeline, model architecture, loss functions, and evaluation metrics of a large‑scale multi‑label image classification system built for 58.com, showing how it improves semantic similarity detection, recommendation, and content moderation across diverse business domains.
Alibaba’s advertising team introduces CREATER, a CTR‑driven text generator that leverages user reviews, aspect control codes, and contrastive fine‑tuning, and CONNA, a non‑autoregressive bundle creator that predicts heterogeneous ad elements with set‑based loss, both delivering substantial online CTR gains and CPC reductions through dynamic creative optimization.
This article surveys recent deep learning approaches for matching in search and recommendation systems, presenting a unified view of matching, categorizing methods into representation learning and matching function learning, and detailing model architectures from input to output layers, while highlighting broader applications such as QA and image captioning.
The 2021 Taobao New‑Product team boosted e‑commerce recommendation by redesigning the coarse‑ranking stage with a dual‑tower DSSM, low‑cost feature‑crossing, NOVA attention and multi‑task distillation from a fine‑ranking teacher, delivering up to +30‰ GAUC gain and 3‑5 % online CTR and click improvements.
The article surveys the re‑ranking stage of modern recommendation pipelines, detailing its architecture after recall and precise ranking, and examining how shuffling and diversity improve user experience, while multi‑task fusion, context‑aware learning‑to‑rank, real‑time online learning, and traffic‑control strategies balance accuracy, efficiency, and business responsiveness.
Meituan's To‑Store Comprehensive Knowledge Graph (GENE) centralizes user demand nodes across diverse local‑life industries, detailing its multi‑layered construction, data mining pipelines, model‑driven entity and relationship extraction, and practical applications in search, recommendation, and intelligent display, while outlining future expansion plans.
This article introduces Meituan's on‑site comprehensive knowledge graph, detailing its multi‑layer design, data‑driven construction pipeline, challenges of diverse user demands and industry complexity, and showcases practical applications in search, recommendation, intelligent display, as well as future expansion plans.
This article presents a comprehensive overview of NetEase Cloud Music's advertising system, detailing its architecture, core challenges, CTR and CVR prediction models, feature engineering, model evolution from LR to deep learning, user vector modeling, and practical recommendations for improving ad performance.
Alibaba’s home‑decor platform introduced a two‑stage cold‑start pipeline—Uniform Guarantee and Boost Amplification—combined with a Wide & Deep content‑potential model that predicts new item popularity, dramatically reducing exposure latency, boosting click‑through rates by ~8 % and overall exposure by 13 %.
This article details JD's heuristic question‑answering framework for intelligent customer service, covering its pre‑consultation prediction, in‑consultation associative input, post‑consultation recommendation modules, underlying algorithms, deployment results, and future enhancement directions.
The framework tags short videos by fusing text, image and audio‑video features through specialized extraction, classification, generative and retrieval modules, then ranking candidates with a multimodal BERT model, delivering accurate, business‑specific tags that boost recommendation, search and advertising.
This talk presents Alibaba's travel intent prediction system, detailing the unique challenges of low‑frequency, multi‑source travel behavior, the multi‑granular CNN and time‑attention model architecture, experimental comparisons with baselines, and how integrated user interest modeling improves recommendation performance.
