CTR預估系列一覽表

01-28

本篇作為目錄頁，希望可以幫助到大家快速找到希望了解的章節；
感謝大家一直以來的支持和幫助，有任何意見或建議可留言或私信等交流討論；
CTR預估作為一個生命力頑強且不斷發展的領域，歡迎各位老師指點雅正。

關於CTR預估系列：

本系列脫胎於外講的系列講稿，在組織文章方面基於如下幾點考慮：

我們在工作中發現，很多同學對於基礎模型的一些細節並不是非常了解，而這些細節可能會影響CTR策略的實施和調優。所以，我們在對主流演算法的介紹中會穿插一些基礎要點的理解與推導、內涵與外延；
有些知識點本身對於CTR預估的應用關係不大，但對於完整地理解問題有幫助。對於這部分相對不緊密的部分，我們會用各類「Aside篇」來說明表示；
CTR預估領域本身是一個發展中的領域，創新點眾多；對於相對成熟的細分領域，我們會盡量概括並給出綱要；部分概括和綱要會更重視整體的流轉，為便於理解而酌情放棄一些細節。
一些子領域中，各家觀點相左；這裡會列舉各家觀點，對於相對重要的部分，會給出筆者這邊的實驗結果。

本系列的文章預計會寫40~60篇，從模型側、特徵側、特徵工程、評估、工程&並行化、監控、問題追蹤等角度相對詳細的闡述CTR預估的各個方面。

CTR 系列的框架和目錄：

0. 問題描述和主要解法

CTR預估[一]: Problem Description and Main Solution

1. 模型側

模型側總圖

Logistic Regression

Naive LR及LR和統計的關係
LR的正則化
LR的Bias及其運用
LR的Model擴展-MLR

Factorization Machine

FM:理論(margin,objective)和實踐
FM的Model擴展-FFM/BFM/SFM(待填坑)

GBDT

GBDT: Preliminary - bagging&boosting, bias&variance
GBDT: Preliminary - 參數空間優化和函數空間優化
GBM和XgBoost
Aside: Random Forest

GBDT Encoder

GBDT Encoder

Deep CTR(待填坑)
Online Learning(待填坑)
Reinforcement Learning(待填坑)

2. 特徵工程

(待填坑)

3. 特徵側

(待填坑)

4. 評估

(待填坑)

5. Model Debug, Monitor and Online Predicting

(待填坑)

Reference (整理ing.)

Papers

[LR-CTR] Predicting Clicks- Estimating the Click-Through Rate for New Ads by _Microsoft_2007_WWW
[MLR]Learning Piece-wise Linear Models from Large Scale Data for Ad Click Prediction
[FM]Factorization Machines
[FM-FTRL]Factorization Machines with Follow-The-Regularized-Leader for CTR prediction in Display Advertising
[FM-FFM]Field-aware Factorization Machines for CTR Prediction
[FM-FFM]Field-aware Factorization Machines in a Real-world Online Advertising System
[FM-BFM] Bayesian Factorization Machines
[FM-SFM] Sparse Factorization Machines for Click-through Rate Prediction
[GBDT Encoder]Practical Lessons from Predicting Clicks on Ads at Facebook
[FE]Position-Normalized Click Prediction in Search Advertising.
[FE]Click Through Rate Estimation for Rare Events in Online Advertising
[FE]SFP-Rank: Significant Frequent Pattern Analysis for Effective Ranking
[GBDT-GBM]greedy function approximation a gradient boosting machine
[GBDT-XgBoost]XGBoost: A Scalable Tree Boosting System
[GBDT-fastRGF]Learning Nonlinear Functions Using Regularized Greedy Forest
[DNN-Deep CTR]Deep CTR Prediction in Display Advertising
[DNN-Deep FM] DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
[DNN-WnD]Wide & Deep Learning for Recommender Systems
[DNN-FNN]Deep Learning over Multi-field Categorical Data
[Feature]Image Feature Learning for Cold Start Problem in Display Advertising
[Feature]Multimedia Features for Click Prediction of New Ads in Display Advertising
[Feature]The Impact of Visual Appearance on User Response in Online Display Advertising
[Feature]Color Harmonization
[Feature]Measuring colourfulness in natural images
[Feature]Natural color image enhancement Natural color image enhancementand evaluation algorithm based on and evaluation algorithm based onhuman visual system human visual system, 2006

Blogs

Lazy Sparse Stochastic Gradient Descent for Regularized Mutlinomial Logistic Regression
Regularized Regression A Bayesian point of view
Logistic Regression and Odds Ratio：Logistic回歸分析和比值比
FM：FM lecture by CMU
Field-aware Factorization Machines
深入FFM原理與實踐
程序化廣告交易中的點擊率預估
機器學習中的數據清洗與特徵處理綜
用戶在線廣告點擊行為預測的深度學習模型
Deep Learning over Multi-field Categorical Data
第四範式聯合創始人陳雨強：機器學習在工業應用中的新思考
kaggle-2014-criteo Idiot』s
CTR預估中GBDT與LR融合方案

Books

The Elements of Statistical Learning