This article explains the exponential weighted moving average (EWMA) as a practical time‑series approximation method, detailing its motivation, recursive formula, weight decay behavior, typical beta values, and a bias‑correction technique that improves early‑stage estimates.
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.
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 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 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.
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 presentation introduces two recent works from Ant Group that improve causal inference by explicitly using historical control data to reduce selection bias and by fusing heterogeneous multi‑source data, describing the GBCT and WMDL methods, their theoretical foundations, experimental results, and practical applications in finance.
This article explains the architecture of a video search ranking system that combines a deep encoding network, multi‑task expert networks, and a bias‑correction module to jointly optimize relevance, click‑through rate, and watch time for streaming platforms.
This article presents the practical evolution of NetEase Yanxuan's recommendation pipeline, covering background, multi‑objective and cross‑domain modeling, bias correction, loss function enhancements, long‑term value strategies, and multi‑scene modeling, with experimental results and a Q&A session.
CausalMTA introduces a causal‑inference‑based multi‑touch attribution framework that eliminates static user‑attribute bias through variational recurrent auto‑encoder re‑weighting and dynamic preference bias via a gradient‑reversal‑layer‑augmented RNN, producing an unbiased conversion predictor whose Shapley‑value credits outperform existing methods on synthetic, Criteo, and Alibaba datasets.
