Computational Advertising

Dr. Liu Peng is senior director and chief architect of business products at Qihoo 360. He isalso responsible for product and engineering for monetization of 360. After receiving hisPhD from Tsinghua University in 2005, he joined Microsoft Research Asia and studiedcutting-edge artificial intelligence technologies. In 2009, he participated in the founding ofYahoo! Labs Beijing as a senior scientist. He was also chief scientist of MediaV. Dr. LiuPeng is devoted to products and technologies related to big data and computationaladvertising. His public online course “computational advertising” has attracted more than30,000 students on Netease.com, and has been adopted as a basic training material inmany related companies. Moreover, this course has been selected by Peking University,Tsinghua University and Beihang University for their graduates.Wang Chao received his master’s degree from Peking University, and then worked atWeibo and Autohome’s advertising department for some years. He is now a tech leader inthe query recommendation group at Baidu’s portal search department. His work focuses onmachine learning algorithms in computational advertising, and he has won 7th place among718 participants in “predict click-through rates on display ads” organized by Kaggle andCriteo. He is also interested in contributing code for open source machine learning toolssuch as xgboost.

• ContentsFigures, xxiTables, xxviiForeword, xxixPreface (1), xxxviiPreface (2), xxxixPreface (3), xliAuthors, xliiiPART 1 Market and Background of Online Advertising 1CHAPTER 1 ■ Overview of Online Advertising 31.1 FREE MODE AND CORE ASSETS OF THE INTERNET 41.2 RELATIONSHIP BETWEEN BIG DATA AND ADVERTISING 51.3 DEFINITION AND PURPOSE OF ADVERTISING 81.4 PRESENTATION FORMS OF ONLINE ADVERTISING 101.5 BRIEF HISTORY OF ONLINE ADVERTISING 18CHAPTER 2 ■ Basis for Computational Advertising 252.1 ADVERTISING EFFECTIVENESS THEORY 262.2 TECHNICAL FEATURES OF THE INTERNET ADVERTISING 292.3 CORE ISSUE OF COMPUTATIONAL ADVERTISING 302.3.1 Breakdown of Advertising Return 322.3.2 Relationship between Billing Models and eCPM Estimation 332.4 BUSINESS ORGANIZATIONS IN THE ONLINE ADVERTISINGINDUSTRY 362.4.1 Interactive Advertising Bureau 372.4.2 American Association of Advertising Agencies 382.4.3 Association of National Advertisers 38PART 2 Product Logic of Online Advertising 39CHAPTER 3 ■ Overview of Online Advertising Products 413.1 DESIGN PHILOSOPHY FOR COMMERCIAL PRODUCTS 433.2 PRODUCT INTERFACE OF ADVERTISING SYSTEM 443.2.1 Demand-Side Management Interface 443.2.2 Supply-Side Management Interface 473.2.3 Multiple Forms of Interface between Supply and Demand Sides 48CHAPTER 4 ■ Agreement-Based Advertising 514.1 AD SPACE AGREEMENT 524.2 AUDIENCE TARGETING 534.2.1 Overview of Audience Targeting Technologies 544.2.2 Audience Targeting Tag System 574.2.3 Design Principles for Tag System 594.3 DISPLAY QUANTITY AGREEMENT 604.3.1 Traffic Forecasting 614.3.2 Traffic Shaping 614.3.3 Online Allocation 624.3.4 Product Cases 634.3.4.1 Yahoo! GD 63CHAPTER 5 ■ Search Ad and Auction-Based Advertising 655.1 SEARCH AD 675.1.1 Products of Search Advertising 675.1.2 New Forms of Search Ads 705.1.3 Product Strategy of Search Advertising 735.1.4 Product Cases 765.2 POSITION AUCTION AND MECHANISM DESIGN 795.2.1 Market Reserve Price 805.2.2 Pricing Problem 815.2.3 Squashing 835.2.4 Myerson Optimal Auction 845.2.5 Examples of Pricing Results 855.3 AUCTION-BASED ADN 855.3.1 Forms of ADN Products 865.3.2 Product Strategy for ADN 885.3.3 Product Cases 895.4 DEMAND-SIDE PRODUCTS IN AUCTION-BASED ADVERTISING 905.4.1 Search Engine Marketing 905.4.2 Trading Desk 915.4.3 Product Cases 915.5 COMPARISON BETWEEN AUCTION-BASED ANDAGREEMENT-BASED ADVERTISING 93CHAPTER 6 ■ Programmatic Trade Advertising 956.1 RTB 976.1.1 RTB Process 986.2 OTHER MODES OF PROGRAMMED TRADE 1006.2.1 Preferred Deal 1006.2.2 Private Marketplace 1016.2.3 Programmatic Direct Buy 1026.2.4 Spectrum of Advertising Transactions 1036.3 AD EXCHANGE 1046.3.1 Product Samples 1046.4 DEMAND-SIDE PLATFORM 1056.4.1 DSP Product Strategy 1066.4.2 Bidding Strategy 1066.4.3 Bidding and Pricing Processes 1086.4.4 Retargeting 1086.4.5 Look-Alike 1116.4.6 Product Cases 1126.5 SUPPLY-SIDE PLATFORM 1136.5.1 SSP Product Strategy 1146.5.2 Header Bidding 1156.5.3 Product Cases 117CHAPTER 7 ■ Data Processing and Exchange 1197.1 VALUABLE DATA SOURCES 1207.2 DATA MANAGEMENT PLATFORM 1237.2.1 Tripartite Data Partitioning 1237.2.2 First-Party DMP 1237.2.3 Third-Party DMP 1247.2.4 Product Cases 1257.3 BASIC PROCESS OF DATA TRADING 1297.4 PRIVACY PROTECTION AND DATA SECURITY 1317.4.1 Privacy Protection 1317.4.2 Data Security in Programmatic Trade 1347.4.3 General Data Protection Regulations 136CHAPTER 8 ■ News Feed Ad and Native Ad 1398.1 STATUS QUO AND CHALLENGES IN MOBILE ADVERTISING 1408.1.1 Characteristics of Mobile Advertising 1418.1.2 Traditional Creative of Mobile Advertising 1428.1.3 Challenges in Front of Mobile Advertising 1448.2 NEWS FEED AD 1468.2.1 Definition of News Feed Ad 1468.2.2 Key Points about News Feed Ad 1498.3 OTHER NATIVE AD-RELATED PRODUCTS 1508.3.1 Search Ad 1508.3.2 Advertorial 1518.3.3 Affiliate network 1518.4 NATIVE ADVERTISING PLATFORM 1518.4.1 Native Display and Native Scenario 1528.4.2 Scenario Perception and Application 1538.4.3 Product Placement Native Ad 1548.4.4 Product Cases 1578.5 NATIVE AD AND PROGRAMMATIC TRADE 161PART 3 Key Technologies for Computational Advertising 163CHAPTER 9 ■ Technological Overview 1659.1 PERSONALIZED SYSTEM FRAMEWORK 1669.2 OPTIMIZATION GOALS OF VARIOUS ADVERTISING SYSTEMS 1679.3 COMPUTATIONAL ADVERTISING SYSTEM ARCHITECTURE 1699.3.1 Ad Serving Engine 1699.3.2 Data Highway 1729.3.3 Offline Data Processing 1729.3.4 Online Data Processing 1739.4 MAIN TECHNOLOGIES FOR COMPUTATIONALADVERTISING SYSTEM 1749.5 BUILD A COMPUTATIONAL ADVERTISING SYSTEM WITHOPEN SOURCE TOOLS 1759.5.1 Web Server Nginx 1769.5.2 ZooKeeper: Distributed Configuration and ClusterManagement Tool 1789.5.3 Lucene: Full-Text Retrieval Engine 1799.5.4 Thrift: Cross-Language Communication Interface 1799.5.5 Data Highway 1809.5.6 Hadoop: Distributed Data-Processing Platform 1819.5.7 Redis: Online Cache of Features 1829.5.8 Strom: Stream Computing Platform Storm 1829.5.9 Spark: Efficient Iterative Computing Framework 183CHAPTER 10 ■ Fundamental Knowledge 18510.1 INFORMATION RETRIEVAL 18610.1.1 Inverted Index 18610.1.2 Vector Space Model 18910.2 OPTIMIZATION 19010.2.1 Lagrange Multiplier and Convex Optimization 19110.2.2 Downhill Simplex Method 19210.2.3 Gradient Descent 19310.2.4 Quasi-Newton Methods 19510.2.5 Trust Region Method 19910.3 STATISTICAL MACHINE LEARNING 20110.3.1 Maximum Entropy and Exponential Family Distribution 20210.3.2 Mixture Model and EM Algorithm 20410.3.3 Bayesian Learning 20610.4 DISTRIBUTED OPTIMIZATION FRAMEWORK FORSTATISTICAL MODEL 21010.5 DEEP LEARNING 21110.5.1 DNN Optimization Methods 21210.5.2 Convolutional Neural Network 21410.5.3 Recursive Neural Network 21510.5.4 Generative Adversarial Nets 217CHAPTER 11 ■ Agreement-Based Advertising Technologies 21911.1 ADVERTISING SCHEDULING SYSTEM 22011.1.1 Scheduling and Mixed Ad Serving 22011.2 GD SYSTEM 22111.2.1 Traffic Forecasting 22211.2.2 Frequency Capping 22411.3 ONLINE ALLOCATION 22711.3.1 Online Allocation Problem 22811.3.2 Examples of Online Allocation Problems 23011.3.3 Limit Performance Analysis 23211.3.4 Practical Optimization Algorithms 23311.4 HEURISTIC ALLOCATION PLAN HWM 240CHAPTER 12 ■ Audience-Targeting Technologies 24512.1 CLASSIFICATION OF AUDIENCE TARGETING TECHNOLOGIES 24612.2 CONTEXTUAL TARGETING 24812.2.1 Near-Line Crawling System 24912.3 TEXT TOPIC MINING 25012.3.1 LSA Model 25012.3.2 PLSI Model 25112.3.3 LDA Model 25212.3.4 Word Embedding (Word2vec) 25312.4 BEHAVIORAL TARGETING 25512.4.1 Modeling Problem for Behavioral Targeting 25512.4.2 Feature Generation for Behavioral Targeting 25712.4.2.1 Tagging Methods for Various Behaviors 26012.4.3 Decision-making Process for Behavioral Targeting 26112.4.4 Evaluation of Behavioral Targeting 26212.5 PREDICTION OF DEMOGRAPHICAL ATTRIBUTES 26412.6 DATA MANAGEMENT PLATFORM 266CHAPTER 13 ■ Auction-Based Advertising Technologies 26713.1 PRICING ALGORITHMS IN AUCTION-BASED ADVERTISING 26813.2 SEARCH AD SYSTEM 27013.2.1 Query Expansion 27213.2.2 Ad Placement 27413.3 ADN 27513.3.1 Short-Term Behavior Feedback and Stream Computing 27513.4 AD RETRIEVAL 27813.4.1 Boolean Expression 27913.4.2 Relevance Retrieval 28313.4.3 DNN-Based Semantic Modeling 28813.4.4 ANN Semantic Retrieval 292CHAPTER 14 ■ CTR Prediction Model 30114.1 CTR PREDICTION 30214.1.1 CTR Basic Model 30214.1.2 LR Model-Based Optimization Algorithm 30314.1.3 Correction of CTR Model 31214.1.4 Features of CTR Model 31314.1.5 Evaluation of CTR Model 31914.1.6 Intelligent Frequency Capping 32114.2 OTHER CTR MODELS 32214.2.1 Factorization Machines 32214.2.2 GBDT 32314.2.3 Deep Learning-Based CTR Model 32414.3 EXPLORATION AND UTILIZATION 32614.3.1 Reinforcement Learning and E&E 32714.3.2 UCB 32914.3.3 Contextual Bandit 329CHAPTER 15 ■ Programmatic Trade Technologies 33115.1 ADX 33215.1.1 Cookie Mapping 33415.1.2 Call-out Optimization 33615.2 DSP 33815.2.1 Customized User Segmentation 34015.2.1.1 Look-Alike Modeling 34115.2.2 CTR Prediction in DSP 34215.2.3 Estimation of Click Value 34315.2.4 Bidding Strategy 34415.3 SSP 34515.3.1 Network Optimization 346CHAPTER 16 ■ Other Advertising Technologies 34716.1 CREATIVE OPTIMIZATION 34816.1.1 Programmatic Creative 34916.1.2 Click Heat Map 35016.1.3 Trend of Creative 35116.2 EXPERIMENTAL FRAMEWORK 35316.3 ADVERTISING MONITORING AND ATTRIBUTION 35416.3.1 Ad Monitoring 35516.3.2 Ad Safety 35616.3.3 Attribution of Advertising Performance 35716.4 SPAM AND ANTI-SPAM 35916.4.1 Classification of Spam Methods 35916.4.2 Common Ad Spam Methods 36016.5 PRODUCT AND TECHNOLOGY SELECTION 36616.5.1 Best Practices for Media 36716.5.2 Best Practices for Advertisers 37016.5.3 Best Practices for Data Providers 372PART 4 Terminology and Index 375REFERENCES, 381INDEX, 387