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.
CAIG leverages a multimodal large language model, a novel reward model, and product‑centered preference optimization to generate ad images that maximize click‑through rate, achieving state‑of‑the‑art performance in both online and offline evaluations.
This guide shows how to map a low‑level containerd task ID to the corresponding Kubernetes Pod using ctr, kubectl, and jq commands, then provides step‑by‑step debugging actions such as pod inspection, log retrieval, and resource checks to quickly resolve container issues.
The paper presents CAIG, a CTR‑driven advertising image generation pipeline that pre‑trains a multimodal LLM on e‑commerce data, trains a reward model on CTR‑labeled image pairs, and fine‑tunes generation via product‑centric preference optimization, achieving state‑of‑the‑art online and offline performance.
An in‑depth comparison of Docker and containerd explains their architectural roles in Kubernetes, covering lifecycle management, logging, configuration, stream services, CNI networking, command equivalents, and how tools like crictl and ctr replace Docker CLI functions.
This guide walks through moving a Docker‑Compose‑defined web front‑end and MySQL back‑end to a Kubernetes cluster using containerd as the runtime, covering essential tools, conversion commands, container‑keep‑alive tricks, external service access, and common troubleshooting steps.
In a Kubernetes cluster the kube-scheduler pod may stay pending due to a container-name reservation conflict; this guide shows how to locate the offending container with ctr, delete it, restart kubelet, and verify that the pod returns to Running state.
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 reviews the evolution of large‑model recommendation techniques, analyzes the specific demands and obstacles of health‑focused e‑commerce, and details JD Health's practical implementations—including LLM‑enhanced recall, deep item‑to‑item models, and scaling‑law‑driven CTR improvements—while discussing open research questions and future directions.
By fusing sparse‑feature operators, enabling multi‑stream execution, consolidating data copies, and merging inference batches, iQIYI reduced GPU CTR model latency to CPU‑level, boosted throughput over sixfold, and cut operational costs by more than 40%, overcoming launch‑overhead bottlenecks.
The article introduces STAN, a multi‑task recommendation framework that leverages user lifecycle segmentation to jointly optimize CTR, stay‑time, and CVR, detailing the business context, key challenges, solution architecture, offline and online evaluations, and future research directions.
This article reviews Shopee and Tsinghua University's latest work on multi‑task CVR optimization, introducing the Click‑aware Structure Transfer with Sample Weight Assignment (CSTWA) method, which tackles knowledge sharing and conflict between CTR and CVR through a three‑part architecture, and demonstrates its superior performance on industrial and public datasets.
This article presents Shopee's calibration‑based meta‑learning approach for personalized click‑through‑rate prediction in live streaming, detailing business context, modeling goals, model evolution from Calibration4CVR to CBMR, EmbCB and MlpCB optimizations, and multi‑task and multi‑scene extensions that achieve significant AUC and business metric improvements.
This article introduces STAN, a user‑lifecycle‑aware multi‑task recommendation model proposed by Shopee that refines CTR, CVR, and stay‑time predictions by identifying and tracking user states, demonstrates offline gains on Shopee and public datasets, and reports online improvements in click‑through, dwell‑time, and order metrics.
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 outlines the challenges of multi-task and multi-scenario modeling for large-scale C-end services, reviews key industry approaches such as Shared-Bottom, MMoE, PLE, ESMM, and LHUC, and details Zhaozhuan’s own EPNET-based solution that improved click-through and conversion rates.
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.
This article describes the Hybrid Embedding solution proposed by Ant AI Infra to address storage, resource, and feature‑governance challenges of massive sparse CTR models, detailing its multi‑layer storage design, KV‑based parameter server, and performance gains in large‑scale recommendation systems.
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 introduces a neuron‑level shared multi‑task learning framework that jointly estimates click‑through rate (CTR) and conversion rate (CVR), discusses the background and advantages of multi‑task learning, reviews classic shared‑bottom models, describes the proposed pruning‑based architecture, and presents experimental results demonstrating its effectiveness in large‑scale recommendation systems.
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.
This article explains how containerd works as a high‑level container runtime and demonstrates practical usage of its three command‑line clients—ctr, nerdctl, and crictl—for pulling images, managing containers, debugging Kubernetes pods, and performing low‑level runtime operations.
This guide explains what containerd is, how to manage images and containers with its command‑line clients ctr, nerdctl, and crictl, and provides practical examples of pulling, running, inspecting, and cleaning up containers in a Linux environment.
The MetaSpore team detailed a three‑pronged optimization—sparse model tuning, training‑pipeline acceleration, and low‑level framework tweaks—that boosted DeepRec CTR model training efficiency by over three times without sacrificing AUC, securing first place in the global AI competition.
Bayesian hierarchical calibration applies a lightweight, interpretable Bayesian GLM with variational inference to correct pre‑ and post‑click scoring biases, using risk‑aware objectives that reduce calibration error by up to 66%, lift revenue by 5%, and cut conversion costs, while handling cold‑start and dimension‑wise sparsity in online advertising.
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+.
The study reveals that industrial deep click‑through‑rate models often overfit dramatically after the first training epoch—a “one‑epoch phenomenon” caused by the embedding‑plus‑MLP architecture, fast optimizers, and highly sparse features, with performance dropping sharply unless sparsity is reduced or training is limited to a single pass.
This article reviews the evolution of advertising targeting technology, recounts its historical impact, analyzes the underlying machine‑learning models—from early Python classifiers to XGBoost‑Spark, DNN, and attention‑based wide‑deep systems—and discusses why the technique is now waning while outlining possible future integrations with large‑scale recall and cost‑aware optimization.
This article examines how search advertising can be seamlessly blended with user queries by balancing relevance and revenue, reviewing the evolution from portal indexing to recommendation systems, and detailing Baidu's Mobius framework that jointly optimizes relevance, CTR, and eCPM in a unified pipeline.
This article presents an adversarial adaptive framework that aligns source and target domains to address domain shift and severe data imbalance in cold-start cross-domain recommendation, demonstrating significant CTR and CVR performance gains when combined with various state‑of‑the‑art single‑domain models.
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.
This article reviews the evolution of click‑through rate (CTR) prediction models from early logistic regression and factorization machines to deep neural networks, introduces the data‑interaction based RIM (Retrieval & Interaction Machine) architecture with its search and prediction modules, and presents extensive experimental comparisons and future research directions.
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 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 reviews several widely used click‑through‑rate (CTR) prediction models—including Logistic Regression, XGBoost, Factorization Machines, Wide & Deep, DeepFM, DCN, xDeepFM, and AFM—providing their principles, advantages, disadvantages, and links to TensorFlow implementations for quick reuse and deeper understanding.
Alibaba Mama’s brand advertising value system evolves from instant CTR to deep CVR and causal uplift modeling, employing focal loss, multi‑task training, GAN‑based uplift, enriched user‑sequence and UID embeddings, which together improve conversion lift, QINI, and interaction metrics while mitigating exposure bias and delayed feedback.
The article analyzes the shortcomings of current click‑through‑rate benchmarks, explains why leaderboards are valuable, and proposes concrete criteria—including online evaluation, sequential test data, leakage prevention, and read‑only submissions—to build a more realistic and robust CTR benchmarking platform.
The article presents JD's technical evolution for advertising ranking, covering technology selection for recommendation ad sorting, context‑aware CTR re‑estimation, reinforcement‑learning‑based sequence optimization, and a session‑level auction mechanism that together improve monetization efficiency and long‑term user value.
This article presents a comprehensive overview of 58's local service recommendation system, detailing the characteristics of its recommendation scenarios, the evolution of tag and post recommendation pipelines, the underlying deep‑learning models such as Bi‑LSTM, ATRank, DeepFM and ESMM, and outlines future research directions.
This article presents a comprehensive overview of 58 Local Service's recommendation system, detailing the diverse recommendation scenarios, challenges such as information homogeneity and complex user structures, the multi‑stage recall and ranking pipelines, model evolutions from statistical methods to deep learning, and future work to improve data quality and model efficiency.
The 2021 review of Alibaba’s Mama Search Advertising estimation platform details advances in model deepening—such as hash‑based embedding compression, adaptive dynamic parameters and graph neural networks—model interaction via a multi‑stage cascade with ranking distillation and oracle bias, and system efficiency gains from HPC training, mixed‑precision, multi‑hash embeddings, and fp16 quantization that deliver roughly a thirty‑fold speed‑up.
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 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.
The article details Alibaba’s low‑carbon CTR model slimming, showing how binary‑code hash embeddings compress massive feature volumes while the Adaptive‑Masked Twins‑based Layer dynamically reduces embedding dimensions, together cutting storage and compute, lowering collisions, and preserving accuracy for large‑scale search advertising.
Meituan accelerates CTR model inference by fusing operators with TVM, optimizing CPU‑GPU data transfers, manually tuning high‑frequency subgraphs, and dynamically offloading workloads, achieving up to ten‑fold throughput gains on Tesla T4 GPUs while keeping latency stable and only modestly increasing beyond 128 QPS, though compilation remains slow and large‑model support needs improvement.
This article surveys recent industrial multi‑task learning approaches for recommendation, covering models such as Alibaba's ESMM and ESM2, DUPN, Meituan's deep ranking, Google’s MMoE, YouTube’s multi‑objective system, Zhihu’s ranking, and summarizing their architectures, loss functions, and practical gains.
This article presents a comprehensive overview of Zhihu's recommendation page ranking system, detailing the request flow, ranking evolution from time‑based to deep‑learning models, feature engineering strategies, model architectures such as DNN, DeepFM, DIN, multi‑task learning, and lessons learned for production deployment.
This article presents the mechanisms of Taobao’s detail‑page full‑network distribution, introducing background, scenario description, and a series of algorithmic explorations—including CIDM, DTIN, and Tri‑tower models—that leverage the main product (trigger) to reinforce users’ instant interests, improve recall, coarse‑ranking, and fine‑ranking performance, and achieve notable online metric gains.
The paper introduces PAPERec, a personalized multi‑objective recommendation framework that leverages Pareto‑oriented reinforcement learning to generate user‑specific objective weights, enabling the model to approximate Pareto‑optimal solutions and achieve superior click‑through rate and dwell‑time performance in both offline and online experiments.
This article reviews recent academic and industrial advances in click‑through rate prediction, classifying optimization techniques for the three‑layer CTR architecture—Embedding, Hidden, and Output—while summarizing three SIGIR papers on graph‑based user behavior modeling, explicit semantic cross‑feature learning, and learnable feature selection for pre‑ranking.
This article explains the theoretical foundations of CTR modeling, why click‑through rates are intrinsically unpredictable at the micro level, the simplifying assumptions that make binary classification feasible, and how evaluation metrics like AUC, contradictory samples, theoretical AUC bounds, and calibration affect model performance.
CTR modeling predicts click probabilities despite inherent microscopic randomness, treating each impression as an i.i.d. Bernoulli event and framing the task as binary classification; because data are noisy and imbalanced, evaluation relies on AUC rather than accuracy, with theoretical upper bounds set by feature quality, and calibration is needed to align predicted values with observed frequencies.
This article introduces a personalized re‑ranking model for recommendation systems, explaining the limitations of traditional point‑wise ranking, describing the PRM architecture with input, encoding, and output layers using multi‑head attention and pre‑trained personalization features, and presenting experimental results and future extensions.
This article presents a comprehensive model quality assurance framework for click‑through‑rate (CTR) prediction, detailing the challenges of data and logic inconsistency, defining consistency goals, describing a full‑stack verification pipeline—including online data capture, offline sample alignment, multi‑stage q‑value comparison, and automated reporting—and sharing practical deployment experiences and results.
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.
The article explains why relying on a single metric like click‑through rate is insufficient for recommendation systems, and outlines a comprehensive, multi‑dimensional evaluation framework that combines business indicators, user behavior metrics, and algorithmic performance measures such as recall, precision, and AUC.
Explore the evolution of Containerd from Docker’s early days, its architecture, installation on Ubuntu, configuration nuances, performance benchmarks, and practical usage with the ctr CLI, while also learning how it integrates with Kubernetes, Docker, and tools like Sealos for streamlined container management.
This article introduces Baidu's new multi‑objective recall system (MOBIUS), which integrates relevance and business metrics such as CPM into the recall stage by migrating CTR models to recall, using data augmentation and a teacher‑student framework to improve ad monetization while preserving relevance.
This article presents an in‑depth overview of Weibo's large‑scale machine learning platform, detailing its multi‑layer architecture, development workflow, CTR model evolution, and how Apache Flink is employed for real‑time data processing, sample services, multi‑stream joins, multimedia feature generation, and future roadmap plans.
The article surveys typical challenges in recommendation systems, including ambiguous evaluation metrics, the trade‑off between precise algorithms and user experience, the exploration‑exploitation dilemma, and why offline AUC improvements often lead to online CTR/CPM drops due to data leakage, feature inconsistency, and distribution shifts.
The article explains why recommendation‑system optimization goals must align with business objectives, illustrates this with YouTube’s watch‑time target and Alibaba’s multi‑task CVR/CTR model, and stresses the strategic importance of defining clear goals to guide cross‑team collaboration.
Recent industrial advances in multi-task learning for recommendation systems, including Alibaba's ESMM and DUPN, Meituan's deep ranking, Google's MMoE, YouTube's multi-objective ranking, and Zhihu's multi-goal model, demonstrate how shared embeddings and specialized loss functions improve CTR, CVR, and user engagement metrics.
This article describes Alibaba‑Mama’s research on improving click‑through rate (CTR) prediction by modeling users’ long‑term interests with incremental computation services, memory‑network architectures (MIMN and HPMN), system redesign (UIC and RTP), and extensive offline and online experiments that demonstrate significant GAUC and CTR gains.
This article details how 58.com constructed a large‑scale personalized recommendation platform for its recruitment business, covering business background, user intent modeling, knowledge‑graph and NER techniques, user profiling, multi‑stage recall strategies, ranking model pipelines, serving infrastructure, AB testing, and future research directions.
This article describes how a multi‑task learning framework based on ESMM and attention‑shared embeddings was built to jointly predict click‑through rate, conversion rate, and gross merchandise value in a large e‑commerce platform, addressing data sparsity, bias, and training challenges.
By moving real‑time recommendation computation to users’ devices and using a two‑stage edge‑cloud CTR model to gate Xianyu’s Mario feature, the system cut server requests by 28%, boosted Mario’s click‑through rate by 31% and lifted overall conversion by 10% while preserving user privacy.
This article reviews the evolution of click‑through‑rate (CTR) estimation models for recommendation ranking, covering logistic regression, feature‑engineering tricks, factorization machines, deep neural networks, wide‑and‑deep architectures, and the Deep & Cross Network, while discussing their strengths, limitations, and future research directions.
This article interprets Alibaba’s “Image Matters” paper, explaining how the Deep Image CTR Model (DICM) introduces user‑side visual preference modeling with image embeddings, why traditional Parameter Servers struggle with large image vectors, and how the Advanced Model Server (AMS) compresses embeddings to enable efficient distributed training.
This article presents a comprehensive overview of Zhihu's recommendation page ranking system, covering its request flow, historical ranking evolution, feature design, deep learning models, multi‑task CTR optimization, practical engineering insights, current challenges, and future research directions such as reinforcement learning.
This article explains how short‑video platforms enhance recommendation quality by combining click‑through‑rate models with multi‑objective optimization of watch time and completion rate, using sample reweighting and weighted logistic regression to balance perceived and real relevance while improving offline AUC and online user engagement.
The author, a graduate student and weekly champion of the Tencent advertising algorithm contest, shares practical guidance on selecting mainstream CTR models—including LightGBM, field‑aware factorization machines, and deep learning approaches—while offering tips on feature handling, hyper‑parameter settings, and resource‑efficient implementation.
This article reviews the fundamentals of click‑through‑rate (CTR) and conversion‑rate (CVR) prediction, explains why the problem is challenging due to high‑dimensional sparse features, and summarizes classic and modern modeling approaches—including feature engineering, linear models, factorization machines, GBDT‑LR, and deep neural networks—while providing practical code snippets and useful research links.
This article describes how Meituan‑Dianping leverages deep learning, especially the Wide & Deep model, to improve its recommendation system by addressing business diversity, user context, feature engineering challenges, optimizer and loss function choices, and presents offline and online experimental results showing significant CTR gains.
The article recounts a junior engineer's journey improving a news app's recommendation system, moving from naive click counts to recent clicks, CTR, lower confidence bounds, and advanced multi‑armed bandit techniques like UCB and Thompson Sampling to balance relevance and novelty.
