Amap shares four conference talks detailing how they built trustworthy AI agents for business decisions, covering semantic governance, causal evaluation, evaluation-driven development, and a product architecture shift that lifted accuracy from 50% to 95% by separating deterministic computation from model reasoning.
This article details how to build automated fault localization for 10M QPS systems by unifying entity identities, aligning timestamps, integrating change records, and applying a four-layer engine—anomaly normalization, temporal correlation, topological pruning, and causal scoring—to converge millions of anomalies into verifiable hypotheses while avoiding correlation-causation pitfalls through counterfactual evidence and phased rollout.
This article reviews the paper “Causal Inference with Unstructured Outcomes”, explaining how the authors extend causal analysis from scalar results to text and image data by defining a max‑contrast feature, presenting identification conditions, estimation algorithms, and extensive experiments on formalness, toxicity, image blur, and paired treatment‑result scenarios.
The article showcases three CIKM 2026 papers from Amap—GenMRP for real‑time multi‑route planning, IntSR a deployed generative search‑and‑recommendation system, and FunnelCausalNet for multi‑tier coupon allocation—detailing their motivations, methods, experimental results, and industrial impact.
In 2026, as software delivery accelerates and AI-native applications proliferate, shift-left testing has become a strategic necessity, with four emerging trends—requirement testability modeling, AI-driven contract verification, atomic quality gates, and causal-inference-based impact measurement—reshaping how quality is built into development.
The article explains the difference between correlation and causation, illustrates three mechanisms that create coincident trends, introduces the Third‑Thing Test for hidden confounders, and offers practical questions to avoid common causal‑mistake traps in data‑driven decision making.
The article examines how 2026’s new A/B testing automation paradigm—combining dynamic traffic allocation, real‑time causal modeling, metric‑autonomy systems, and built‑in privacy compliance—dramatically shortens experiment cycles, boosts statistical power, and transforms experimentation from a manual chore into a scalable, trustworthy decision engine.
This article examines the Wan2.1 video diffusion model, identifies its scalability bottlenecks for long and real‑time video generation, and introduces the Self‑Forcing causal framework together with sequence‑parallel and RoPE optimizations that achieve sub‑second latency and up to 1.5× speed‑up on modern GPUs.
This comprehensive survey examines the emerging field of Causal Reinforcement Learning, classifies its core techniques, introduces eleven benchmark environments, evaluates four novel algorithms, and outlines challenges and future research directions for building robust, generalizable, and interpretable AI systems.
Judea Pearl’s Ladder of Causation framework divides reasoning into three levels—association, intervention, and counterfactuals—explaining how conditional probability, the do‑operator, and structural causal models enable moving from mere data correlation to actionable causal insights, with practical criteria like back‑door and front‑door adjustments.
This article compares six popular Python causal inference libraries—Bnlearn, Pgmpy, CausalNex, DoWhy, PyAgrum, and CausalImpact—using the U.S. Census Income dataset to answer whether a graduate degree raises the probability of earning over $50K, and provides detailed code, pros, cons, and results for each tool.
This article explains the concept of heterogeneous treatment effects, outlines how to select dimensions for HTE analysis, describes a Python‑based MVP tool for automated CATE exploration, and showcases a real‑world experiment case where sub‑group insights turned a non‑significant overall result into actionable findings.
This article explains experimental heterogeneity (HTE), clarifies key concepts such as CATE and ITE, discusses why analyzing treatment‑effect variation matters for business, compares statistical and machine‑learning methods, and introduces an open‑source Python tool that automates CATE discovery and reporting.
This article explains the concept of heterogeneous treatment effects (HTE), clarifies related terminology, outlines why HTE analysis matters for product decisions, and walks through dimension selection, statistical and machine‑learning methods—including ANOVA, causal trees, meta‑learners, and double‑machine‑learning—plus a practical MVP tool with code examples and future development directions.
An in‑depth, five‑minute guide compares six popular Python causal inference libraries—Bnlearn, Pgmpy, CausalNex, DoWhy, PyAgrum, and CausalImpact—using the Census Income dataset to illustrate structure learning, parameter estimation, inference, and causal effect validation, highlighting each tool’s strengths, limitations, and ideal use cases.
This article systematically compares six popular Python causal inference libraries—Bnlearn, Pgmpy, CausalNex, DoWhy, PyAgrum, and CausalImpact—using the U.S. Census income dataset to demonstrate how each tool discovers causal effects of education on salary, highlighting their core features, strengths, weaknesses, and suitable scenarios.
This digest summarizes three recent AI papers—DoFlow, Forecast2Anomaly, and ForecastGAN—detailing their causal generative flow model for interventions, a retrieval‑augmented framework for zero‑shot anomaly prediction, and a decomposition‑based adversarial approach that improves multi‑horizon forecasting across diverse datasets.
This digest presents seven recent arXiv papers that explore large‑language‑model‑driven portfolio scoring, hybrid ResNet‑RMT covariance denoising for crypto, LLM‑enhanced financial causal analysis, multilingual news alignment for stock returns, three‑step bubble prediction with news and macro data, multimodal volatility forecasting, and news‑aware reinforcement trading, each with reported performance gains.
Pearl’s do‑calculus provides a mathematical framework to derive intervention effects from causal graphs, enabling researchers to predict how policy changes—such as increased vaccination rates—affect disease incidence, with three core rules guiding causal reasoning, substitution, and counterfactual analysis.
In an interview, Di Xingxing of Autohome details their AI+Data framework—unified lake‑warehouse, intelligent engine, and agent services—that breaks data silos, blends traditional models with LLMs, leverages causal inference and RAG knowledge bases, and uses continuous feedback to build explainable, evolving data agents for accurate sales forecasting, competitive analysis, and end‑to‑end business automation in the automotive industry.
This article presents a unified framework for efficiently estimating causal treatment effects on exponential‑family outcomes, extending target regularization beyond Gaussian assumptions, deriving bias analysis for plug‑in estimators, proposing DR and TMLE‑based estimators, and validating them on synthetic and real datasets.
This article describes a QCon talk that combines causal inference with large language models to build a retrieval‑augmented generation pricing system for e‑commerce, detailing the three‑step algorithm, LLM‑driven modeling challenges, process‑reward tree search, reinforcement‑learning fine‑tuning, and experimental gains in accuracy and speed.
This article presents a data‑fusion framework that combines long‑term observational data and short‑term randomized experiments to identify and estimate long‑term heterogeneous dose‑response curves under continuous treatments and unobserved confounders, using reweighting, optimal transport, and balanced representation learning.
This article describes how integrating causal inference with large language models and Retrieval‑Augmented Generation can automate and explain e‑commerce product pricing, detailing the three‑step workflow, reinforcement‑learning rewards, experimental results, and future directions for end‑to‑end RAG‑LLM training.
At QCon 2025, the author presented a novel approach that integrates causal inference with large language models using Retrieval‑Augmented Generation, process rewards, and tree‑search to generate explainable, accurate e‑commerce pricing recommendations, dramatically improving accuracy from 44% to 74% while cutting inference time to seconds.
This article presents a comprehensive approach that combines causal inference, large language models, and retrieval‑augmented generation to automate e‑commerce price recommendation, detailing the three‑step workflow, challenges across product categories, the RAG architecture, process‑reward‑guided tree search, reinforcement learning refinements, and experimental results showing significant accuracy and speed improvements.
This article explains the challenges of evaluating uplift models for intelligent marketing with multiple discount treatments, reviews existing metrics such as AUUC, Qini, and AUCC, and introduces the RAS‑AUCC metric that aligns offline evaluation with online ROI by sorting samples by marginal ROI and using RCT data.
This article explains the fundamentals of causal inference, distinguishes it from correlation, introduces major theoretical frameworks such as structural causal models and potential outcomes, and demonstrates practical uplift modeling techniques—including meta‑learners, double machine learning, and deep causal networks—through a financial credit‑limit use case.
Kuaishou has had eight cutting‑edge papers accepted at the International Conference on Machine Learning 2025, covering breakthroughs in multimodal emotion modeling, monotonic probability learning, causal effect generalization, cascade ranking, multimodal LLM alignment, ultra‑low‑rate image compression, and visual autoregressive super‑resolution, with links to each work and accompanying code repositories.
This article explains how Tencent Music’s experiment platform automates strategy‑positive subpopulation mining using unified dimension tables, CATE model training, double‑difference estimation, and propensity‑score matching, enabling rapid recommendation‑strategy optimization and data‑driven product decisions.
This article explores the fundamentals and challenges of uplift modeling, explains why unbiased random data are essential, and presents a comprehensive suite of bias‑correction methods—including reweighting, propensity‑score matching, and advanced deep‑learning architectures such as TarNet, CFRNet, and DragonNet—to improve causal effect estimation in marketing and finance applications.
Observational research methods—synthetic control, matching, and Causal Impact—offer powerful alternatives to randomized experiments, enabling businesses like Meituan to assess policy effects despite legal and operational constraints, with detailed principles, applications, advantages, limitations, and practical case studies illustrated throughout the guide.
This article, part of the Trusted Experiment Whitepaper series, introduces quasi‑experimental design and focuses on the Difference‑in‑Differences (DID) method, explaining its principles, evaluation models, parallel‑trend testing, extensions, and a concrete Meituan fulfillment case study illustrating practical implementation.
This article explains what natural experiments are, how they differ from controlled trials, and provides practical steps, classic cases, and analytical methods like DID, RDD, and IV to help readers discover and design credible real‑world experiments.
This article explains the statistical foundations of AB testing, covering the Rubin causal model, SUTVA and randomization assumptions, parameter and confidence‑interval estimation, hypothesis‑testing procedures, and essential limit theorems such as the law of large numbers and the central limit theorem.
This whitepaper introduces the importance of data‑driven A/B testing, outlines its theoretical foundations, practical challenges such as small samples and spillover effects, and presents a structured roadmap—including experiment basics, statistical principles, advanced designs, and SDK implementation—to help practitioners design trustworthy experiments.
This announcement introduces a live session that will dissect two best‑paper award research works from WSDM 2025—one revealing how recommendation models amplify popularity bias through spectral analysis and proposing a lightweight regularizer, and the other presenting a graph disentangle causal model that integrates GNNs with structural causal models to improve causal inference on networked observational data.
Using a year‑long revenue dataset from an online‑education firm, this article examines how description, causal explanation, and statistical modeling together reveal patterns, uncover underlying drivers, and highlight the limits and uncertainties of forecasting future performance.
This article compiles concise overviews of recent e‑commerce research, covering real‑time online learning re‑ranking models, causal inference for user growth, full‑link data lineage, TikTok's data governance and attribution solutions, Volcano Engine's metric management, AI Agent applications on 1688, and XinXuan Group's live‑stream data architecture.
The paper introduces a coupled generative adversarial framework that merges biased observational with unbiased experimental data to create a bias‑free dataset for causal inference, enabling robust treatment‑effect estimation under collider bias from an out‑of‑distribution perspective, and demonstrates superior bias reduction on three public advertising datasets.
This article introduces causal inference concepts, explains structural causal and potential‑outcome frameworks, presents the MACR model for debiasing popularity in recommender systems, and details two experiments conducted on the ZhaiZhai platform along with future research directions.
This article presents an overview of weak supervision machine learning techniques applied to Ant Group’s business scenarios, covering an introduction to weak supervision, challenges of modeling with scarce or noisy labels, detailed methodologies for cross‑domain causal effect estimation, multi‑source noisy label denoising, and real‑world application examples.
This article presents a comprehensive overview of using large language models to automate the full pipeline of social‑science research—from hypothesis generation and agent construction to experiment execution, data collection, and model estimation—illustrated with a simulated auction study and a discussion of future directions.
This article explains metric attribution for internet platforms, covering its definition, a three‑step analytical framework, deterministic and probabilistic methods such as metric decomposition, machine‑learning models with SHAP values, case studies, and a practical tool that guides users through attribution analysis.
This article introduces causal debiasing techniques for Ant Group's marketing recommendation systems, detailing background biases, causal graph analysis, a meta‑learning data‑fusion model (MDI), backdoor‑adjustment methods, extensive experiments on public and internal datasets, and real‑world deployment results.
This article explains the fundamentals of metric attribution, outlines a three‑step framework for identifying, analyzing, and solving metric changes, compares deterministic, probabilistic, and speculative methods, and illustrates the approach with two real‑world case studies using decomposition and machine‑learning techniques.
The article reviews the development of online A/B testing, covering sampling and traffic‑splitting techniques, metric computation improvements, statistical inference advances, and current challenges such as interference, real‑time inference, and large‑scale metric computation, while referencing recent research papers.
This article presents a comprehensive technical overview of weak‑supervision machine learning at Ant Group, covering its fundamentals, cross‑domain causal effect estimation, strategies for scarce or noisy labels, novel framework components, experimental validation, and practical application scenarios.
This article surveys recent advances in applying causal inference to recommender systems, presenting three lines of work—causal embedding for interest‑conformity disentanglement, contrastive learning for long‑term and short‑term interest separation, and adversarial debiasing of duration bias in short‑video recommendation—along with experimental validation and insights.
This article presents a comprehensive overview of weak supervision machine learning techniques applied to Ant Group's business problems, covering theoretical foundations, cross‑domain causal effect estimation, noisy‑label denoising frameworks, experimental results, and practical use cases such as risk modeling and marketing interventions.
This article reviews the role of world (mental) models and causal inference in reinforcement learning, covering their theoretical foundations, model‑based RL frameworks such as Dyna, sample‑efficiency challenges, causal structure learning, distribution correction, dynamics‑reward modeling, and experimental results that demonstrate performance gains across multiple tasks.
This article presents a detailed attribution analysis framework—including background research, a four‑step workflow, Bayesian causal detection, Simpson's paradox handling, and real‑world case studies—demonstrating how data‑driven insights can improve conversion rates and operational efficiency across multiple business lines.
The paper introduces CausalMMM, a variational inference framework that integrates Granger causality and graph neural networks to automatically discover heterogeneous causal structures in marketing mix modeling, enabling more accurate GMV prediction and actionable insights for diverse advertisers.
This article introduces causal inference fundamentals, distinguishes correlation from causation, reviews major methodological streams, and demonstrates how uplift and gain models—implemented with T‑learner, S‑learner, and tree‑based approaches—can be applied to user growth and marketing scenarios, including evaluation metrics and future challenges.
When large‑scale A/B testing is impractical for high‑traffic, socially intensive games, developers can rely on methods such as Difference‑in‑Differences, hypothesis proportion analysis, and differential‑ratio comparison to infer the causal impact of content updates on key performance metrics.
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.
This article presents a novel approach that combines debiased deep learning with double machine learning to estimate and infer average treatment effects across multiple simultaneous online experiments, detailing problem definition, a semi‑parametric theoretical framework, and extensive field‑experiment validation on a large video‑platform dataset.
The DataFun Data Science Summit on May 25 brings together leading experts to present cutting‑edge research on pairwise data causal inference, Retrieval‑Augmented Generation applications, large language model content detection, user growth analytics, and advanced machine‑learning techniques across finance, e‑commerce, and AI domains.
The DataFun Data Science Summit on May 25 will feature Tencent data scientist Li Yilin presenting a comprehensive talk on pairwise‑data A/B experiments, covering unbiased estimation under various randomizations, theoretical analysis, and practical insights for causal inference in network‑driven online experiments.
The DataFun Data Science Summit on May 25 will feature Tencent data scientist Li Yilin presenting a comprehensive overview of paired‑data A/B experiments, covering causal inference challenges, unbiased estimators under various randomization designs, theoretical analysis, and practical insights for network‑based online experiments.
The article explains the necessity of incorporating causal relationships into machine learning, outlines the development of causal science, and details how uplift modeling and causal‑regularized stable learning are applied to marketing and risk control in the banking sector, while also discussing practical challenges and experimental results.
This article examines how game AB experiments can exhibit significant regional differences, outlines a meta‑analysis framework to detect heterogeneity, decomposes its sources into treatment‑effect and distributional factors, and demonstrates how to predict outcomes for unseen regions using machine‑learning models.
This article presents causal inference–based methods to address bias in recommendation systems, covering the transformation of recommendation problems into causal problems, selection bias mitigation through double‑robust and multi‑robust learning, individual treatment effect estimation, and a case study on attention bias in music recommendation.
Structural Equation Modeling (SEM) combines multiple regression analyses to simultaneously assess direct and indirect relationships among observed and latent variables, offering advantages such as handling multiple causal paths, incorporating latent constructs, flexible error modeling, and testing mediation and moderation effects, illustrated with an education‑investment case study.
The presentation surveys recent causal‑inference research for recommendation systems, introducing the DICE framework to separate user interest from conformity, the CLSR model to disentangle long‑term and short‑term preferences, and the DVR approach with WTG metrics to debias short‑video recommendations, demonstrating improved accuracy, fairness, and interpretability.
This presentation reviews recent research on applying causal inference to recommendation systems, covering causal embedding for separating user interest and conformity, contrastive learning for disentangling long‑term and short‑term interests, and a debiasing framework for short‑video recommendation that uses watch‑time‑gain metrics and adversarial learning to mitigate duration bias.
This article explores a novel framework that merges causal inference with genetic algorithms to improve freight pricing strategies, addressing data limitations, bias, and dynamic market conditions, and demonstrates its robustness and effectiveness through extensive offline simulations and real‑world experiments.
This article explores the challenges of estimating hotel occupancy in online booking platforms and presents four comprehensive approaches—background analysis, demand‑function based quantity‑price modeling, causal‑inference modeling, and multi‑scenario joint modeling—highlighting novel models, datasets, and experimental results for dynamic pricing optimization.
This article presents a logistics‑focused case study that leverages causal inference techniques, including uplift modeling and entropy‑balancing with flexible spatiotemporal grids, to enhance marketing strategy efficiency using observational data while addressing industry‑specific technical challenges.
This article examines why A/B experiments often capture only short‑term impacts, explains external and internal factors behind short‑ and long‑term effects, and presents seven industrial methods—including user‑learning models, personalized recommendation adjustments, surrogate metrics, and bias correction—supported by real‑world case studies.
This article explains the concept of mediation effects, outlines how to select mediators, describes step‑by‑step testing procedures—including the classic Baron‑Kenny approach and modern bootstrap methods—and demonstrates the whole workflow with a real‑world economic example.
This article introduces causal learning, explains its distinction from traditional correlation‑based machine learning, outlines its three main parts, discusses the two primary paradigms—learning with known causal graphs and learning via causal discovery—and highlights their advantages, challenges, and recent research directions.
This article explains how causal inference techniques, especially uplift models combined with entropy‑balancing on spatio‑temporal grids, improve marketing strategy efficiency in the logistics sector by addressing confounding bias in observational data.
This article presents Ant Group's research on causal debiasing for recommendation and marketing, covering the background of bias, common bias types, causal graph analysis, two correction approaches—data‑fusion based MDI and back‑door adjustment based DMBR—along with experimental results on public and proprietary datasets and real‑world deployment insights.
This article presents how causal inference methods are applied to Kuaishou’s single‑column short‑video recommendation, covering the platform’s recommendation scenario, model representations, duration bias mitigation, viewing‑time prediction techniques such as D2Q and TPM, experimental results, and future research directions.
This article presents how Ant Group applies causal inference techniques—including confounding bias analysis, double‑difference methods, DiDTree, and shrinkage‑based causal trees—to correct biases in risk‑control scenarios, detailing the theoretical background, algorithmic designs, experimental validation, and practical deployment.
Meituan’s Fulfillment Platform team built a high‑performance distributed causal‑forest framework—named Causal On Spark—that trains hundreds of trees on hundreds of millions of samples within minutes using MapReduce‑based histogram splitting, extensive memory optimizations, Parquet model serving, and novel distributed evaluation metrics, enabling scalable causal inference for pricing, subsidies, and marketing.
This article presents a detailed overview of world models and their role in reinforcement learning, explains how causal inference can enhance model-based RL, discusses sample efficiency challenges, and shares experimental findings and practical insights from recent research and industry applications.
This article reviews the importance of causal inference in medicine, covering historical perspectives on disease causation, epidemiological methods such as Mill's rules and cohort studies, modern techniques like Mendelian randomization, and future research directions in causal graph learning and AI integration.
This article presents Ant Group's causal debiasing techniques for recommendation and marketing, covering bias background, data‑fusion based MDI model, back‑door adjustment methods, experimental results on public and industry datasets, and practical applications in advertising and e‑commerce.
This article explains how the Huolala data‑science team tackles AB‑testing challenges, pre‑experiment differences, observational (non‑AB) studies, and advanced causal‑inference methods such as CACE, heterogeneous treatment effects, mediation modeling, regression discontinuity, and instrumental variables to derive reliable business insights.
This article explains why causal inference tools are essential for industrial growth, outlines data‑flow standards such as randomized controlled trials, discusses model selection including causal forests and policy learning, and describes evaluation, offline simulation, and resource‑constrained optimization for deploying causal models in production.
This article presents a logistics‑focused case study that leverages causal inference techniques, especially uplift modeling combined with entropy‑balancing and flexible spatio‑temporal grid partitioning, to enhance marketing strategy efficiency, address confounding bias, and achieve stable, accurate effect estimation across diverse operational scenarios.
This article examines why A/B experiments often capture only short‑term impacts, categorises external and internal causes of short‑term bias, and presents seven industry‑tested approaches—including user‑learning models, personalized recommendation adjustments, surrogate metrics, and bias correction techniques—to reliably estimate long‑term strategy effectiveness, illustrated with real business cases.
This article examines how two‑sided platforms such as Uber, Lyft, DoorDash and Didi design and evaluate peak‑surcharge experiments, addressing network effects, bias‑variance trade‑offs, time‑space slicing, random‑saturation designs, and continuous bandit‑based testing within an operations‑focused experimental system.
This article reviews the growing interest in causal learning within machine learning, explaining what causal learning is, its advantages over purely correlational methods, and detailing two main paradigms—learning with known causal structures and learning via causal discovery—along with examples, challenges, and future directions.
Huya’s data‑driven experiment platform showcases how causal inference, AB testing, and uplift modeling are applied to advertising, user activation, and growth scenarios, detailing platform evolution, metric design, statistical challenges, and practical solutions such as multi‑test correction, CUPED, RTA, and propensity‑score methods.
This article examines the unique challenges of conducting experiments in the freight bilateral market—covering transaction, pricing, marketing, and product scenarios—and presents a comprehensive technical solution framework that includes cluster traffic splitting, homogeneity assurance, efficient interpretation, and observational study methods.
This article explains how to build a data‑driven metric system—from defining end‑to‑start metrics and combining business and data drivers, to applying statistical analysis, machine‑learning, causal inference, and practical case studies for alerting, diagnosing, and strategizing product performance.
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.
This article presents a comprehensive overview of using instrumental variables for causal inference and causal generalization in machine learning, discussing deep learning limitations, Pearl's causal hierarchy, two‑stage regression, challenges with unobserved confounders, automatic IV generation, and applications in economics and social networks.
This article presents OPPO Andes Smart Cloud's intelligent growth algorithm framework for the smartphone sector, detailing industry background, data and model architecture, four real-world application cases—including AIGC content generation, multimodal recommendation, causal inference, and precise advertising—and summarizing key insights from a technical Q&A session.
This article examines the unique characteristics of freight two‑sided markets, outlines the experimental challenges across transaction, pricing, marketing, and product scenarios, and presents a comprehensive technical framework—including allocation strategies, homogeneity controls, efficient interpretation, and observational study methods—to achieve reliable, actionable insights.
This article explains how to build and apply a data‑driven metric system, covering end‑to‑end design principles, business‑ versus data‑driven approaches, frameworks such as OSM, GSM and HEART, statistical and machine‑learning techniques, causal inference, and practical case studies that illustrate alerting, diagnosis, and strategy deployment in product operations.
The article presents a comprehensive pricing solution for an e‑commerce platform that combines Bayesian statistical modeling, MCMC sampling, and causal inference (including Dragonnet) to achieve controllable, fine‑grained SKU‑level price estimation and optimization.
The DataFunSummit2023 online conference brings together experts from Tencent and Kuaishou to present cutting‑edge research on causal inference for large‑scale A/B testing, including deep‑learning‑based multi‑experiment effect estimation, a distributed causal inference framework (Fast‑Causal‑Inference), and strategies for evaluating long‑term policy impacts.
This article examines the fairness of recommendation systems from both consumer and provider viewpoints, discussing sources of bias, definitions of equality and equity, measurement metrics such as CGF and MMF, and proposes causal embedding models to mitigate unfairness while ensuring sustainable system performance.
The presentation explains how to detect, analyze, and leverage regional variations in overseas A/B test results to make more informed product decisions, using a systematic experimental analysis framework grounded in causal inference and online experimentation methods.
In this talk, senior researcher Dai Quanyu from Huawei Noah's Ark Lab presents his work on applying causal inference to identify and correct various biases in recommendation systems, detailing underlying theoretical frameworks, bias‑mitigation algorithms such as inverse propensity weighting and robust learning, and real‑world product deployments.
This article presents a comprehensive approach to estimating causal effects on clustered data using a double‑difference method, introduces the DiConfounder algorithm built on Rubin Causal Model extensions, details data characteristics, model assumptions, six‑step pipeline, and reports competitive results on the ACIC2022 challenge.
