47 · A01 · Ad products, participants, and transaction flows
Map ad products, participants, transaction forms, and their conflicting objectives.
23 LESSONS · ONE CONNECTED PATH
Advertising
Begin with products and markets, then response prediction, allocation, bidding and feedback.
Before you begin: Start with 31–34; prior auction or CTR-model knowledge is not assumed.
THE LEARNING PATH
Build one idea on the last.
01
Products and markets · A01–A06
Participants, products, billing, contracts, sponsored search, multi-slot auctions and RTB.
48 · A02 · Billing, denominators, and a three-party ledger
Calculate billing and conversion denominators while preserving three-party accounting.
49 · A03 · Contract inventory forecasting and online allocation
Compare contract allocation under arrival-order changes with a tiny offline upper bound.
50 · A04 · Sponsored-search matching and quality ranking
Apply campaign eligibility before quality-weighted sponsored-search ranking.
51 · A05 · Multi-slot GFP, GSP, and VCG
Calculate GFP, GSP, and VCG payments under explicit separable-click assumptions.
52 · A06 · RTB roles, reserves, and censored feedback
Separate RTB role states, reserve prices, first/second-price rules, and censored feedback.
02
Eligibility and candidates · A07–A08
Context, audience expansion, eligibility and inverted candidates through constructed mechanisms.
53 · A07 · Context targeting, expansion, and data boundaries
Study context targeting and expansion while keeping data-use boundaries explicit.
54 · A08 · Ad candidate indexes and constrained matching
Verify constrained ad retrieval against a brute-force oracle and recheck eligibility.
03
Response prediction · A09–A12
Real impressions and a linear baseline → nonlinear FM comparison → constructed conversion delays → calibration audit.
19 · Impressions, labels, missingness, and leakage-aware splitting
Inspect impression labels, missingness, grouped splits, and train-only vocabularies.
20 · A linear CTR baseline and feature engineering
Train numerical-only and categorical-feature logistic CTR baselines on a fixed prefix.
21 · Nonlinearity, feature crosses, and tree capacity
Use XOR to explain interactions, then compare real tree models at two capacities.
55 · A10 · Real CTR: linear versus FM interactions at a fixed budget
Compare real linear and FM training at one controlled budget, retaining FM overfitting.
56 · A11 · CVR, CTCVR, delay, and selection bias
Explain CVR selection and delayed observation using clearly constructed conversions.
22 · Probability scores, reliability, and calibration audit
Inspect proper probability scores, refit a fixed calibration map, and audit uncertainty.
04
Delivery and feedback · A13–A18
Value bidding, feedback pacing, attribution and incrementality, reserve prices, experiment integrity and ledgers.
57 · A13 · Value bidding, budgets, and ROAS constraints
Separate value, target-CPA, and target-ROAS bids from globally optimal budget allocation.
58 · A14 · Budget pacing and feedback control
Compare hard gating and feedback pacing, including drift, underspend, and oscillation.
59 · A15 · Multi-touch attribution and randomized incrementality
Distinguish attribution accounting from incrementality in a known randomized mechanism.
60 · A16 · Reserves, fill rate, and market utility
Sweep reserves on fixed simulated bids and inspect fill, revenue, and bidder utility.
61 · A17 · Ad experiments, bias, and fraud screening
Expose mixture bias, shared-budget interference, and false positives in fraud rules.
62 · A18 · An end-to-end ledger from ad request to settlement
Trace one ad request through eligibility, payment, and an idempotent single-process ledger.
05
Reference · integrated experiments
Original prediction and decision experiments remain as references; the learning sequence starts from the domain.
18 · One system: real click probabilities and simulated ad decisions
Keep real click-probability evidence separate from simulated auction decisions.
23 · Auctions, payment rules, and bidding
Calculate second-price payments and compare bids on the same potential market.
24 · Budget pacing, constraints, and integrated error analysis
Verify per-request budget invariants and separate prediction error from policy assumptions.
No lessons match. Try another concept or clear the search.
TRACE THE IDEAS
Sources & lecture notes
Follow the original material alongside the Lab. Source scope, original language and the boundaries of each teaching unit are explicit.
Liu Peng & Wang Chao · Computational Advertising, 2nd ed.
Only the public table of contents and introduction were checked. A paid textbook, not a free complete lecture set.
Introduction to Computational Communication · Chapter 8
A pinned 96-slide Chinese chapter on products, objectives, games, matching and auctions. Not a complete advertising engineering course.
Nanjing University · Computational Communication and Advertising (2015)
Public syllabus and assignments; not all lecture decks are available. Historical market descriptions are not current market facts.
Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial
164 author slides covering response prediction, bidding, attribution, pacing, targeting and reserve prices. Lab mechanisms are independently authored, not full paper replications.
Weinan Zhang · RTB Paper Index
Further reading and paper discovery; the index itself is not a lecture.
Chapter-to-source map 18 units
A01
Participants, products and market structure
Trace an impression through advertiser, publisher, exchange and user roles.
- Introduction to Computational Communication · Chapter 8 Slides 3–5 and 14–41: products and objectives
- Liu Peng & Wang Chao · Computational Advertising, 2nd ed. Public contents: products and business models
Boundary: A constructed product/role pipeline, not a complete protocol or brand-effect evaluation.
A02
Billing units, metrics and stakeholder objectives
Convert CPM/CPC/CPA units and reconcile spend, value and stakeholder objectives.
- Introduction to Computational Communication · Chapter 8 Slides 14–41: objectives and value
Boundary: A constructed ledger without fees; revenue minus ad spend omits other costs and is not net profit.
A03
Contracts, inventory forecasting and online allocation
Compare offline allocation with online decisions when future inventory is uncertain.
- Liu Peng & Wang Chao · Computational Advertising, 2nd ed. Public contents: contract advertising; inventory mechanism by the Lab
Boundary: Three inventory units, exhaustive allocation versus online greedy, and a historical-mean forecast; not industrial forecasting or an optimality guarantee.
A04
Sponsored search, eligibility and quality ranking
Apply eligibility before quality-aware ranking and diagnose why an ad is excluded.
- Nanjing University · Computational Communication and Advertising (2015) Syllabus: sponsored search and matching
- Introduction to Computational Communication · Chapter 8 Slides 42–94: matching and auctions
Boundary: Keyword OR matching, rule-based eligibility and artificial quality; no real keyword or landing-page quality logs.
A05
GFP / GSP / VCG and multi-slot auctions
Calculate allocations, prices and utilities under explicit auction assumptions.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 21–36: auctions
- Introduction to Computational Communication · Chapter 8 Slides 42–94: games, matching and auctions
Boundary: Homogeneous quality, separable position CTR and no budgets; numerical VCG checks depend on these assumptions.
A06
Programmatic RTB, reserves and market feedback
Trace bids, wins, reserves and feedback through a minimal exchange.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 18–19 and 38: systems
Boundary: Minimal role/feedback states, not OpenRTB. Below-reserve losses and competitive losses are distinguished.
A07
Audience targeting, context, expansion and DMPs
Separate context from identity and inspect audience expansion errors.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 133–148: targeting and expansion
Boundary: Artificial contexts and artificial relevance labels; no real profiles, cross-domain DMP or measured expansion gain.
A08
Ad candidate retrieval and constraints
Compare constrained inverted candidates with an exhaustive oracle.
- Liu Peng & Wang Chao · Computational Advertising, 2nd ed. Public contents: ad retrieval; small index implemented by the Lab
Boundary: A tiny inventory, inverted candidates and exhaustive oracle; no distributed serving or concurrent budget consistency.
A09
Real impressions, splits and linear CTR
Inspect labels, leakage boundaries and the contribution of sparse features.
19 · Impressions, labels, missingness, and leakage-aware splitting20 · A linear CTR baseline and feature engineering
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 39–69: response prediction
Boundary: Real sampled impressions, anonymous fields and feature-group partitioning; not a user/time split or population CTR estimate.
A10
Feature interactions and nonlinear response models
Train FM and a linear control on the same real sample, retaining overfitting results.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF page 58: factorization machines
Boundary: 10k real impressions, categorical fields, one seed and eight epochs. Selection-only development diagnostics preserve FM overfitting.
A11
Conversions, delayed feedback and selection
Explain observation windows, delayed labels and conversion selection bias.
- Liu Peng & Wang Chao · Computational Advertising, 2nd ed. Public contents: CTR/CVR; delayed-feedback exercise by the Lab
Boundary: Explicitly constructed conversions and delays; Criteo clicks are not relabeled as real conversions. No real CVR/ESMM gain.
A12
Calibration, proper scores and segment audits
Use separate calibration/audit partitions and compare proper scores and reliability.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 39–69: response prediction; calibration audit by the Lab
Boundary: Original separate calibration/audit and negative results retained. Unknown sampling priors prevent claims about online calibration.
A13
Bid objectives, bidding functions and value optimization
Distinguish expected value, budget feasibility and ROAS constraints.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 71–91: bidding optimization
Boundary: A fixed constructed second-price market with known expected conversion values. Hard budgets and ROAS thresholds are not globally optimal.
A14
Budget pacing and feedback control
Compare pacing responses under fixed market conditions and changing demand.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 118–131: budget pacing
Boundary: Integral error feedback, not a complete PID controller. Three scenarios expose value/spend tradeoffs, not guaranteed optimal spend.
A15
Attribution and incrementality
Separate multi-touch bookkeeping from causal incrementality.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 94–116: attribution
Boundary: Constructed paths and known random assignment; no real attribution ground truth, propensities or IPS/OPE result.
A16
Reserve prices and supply–demand strategies
Compare fill, revenue and utility under an explicit bid distribution.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 150–163: reserve prices
Boundary: Fixed U(0,2) bids and a common second-price reserve. Strategic long-term responses are omitted; no real reserve recommendation.
A17
Experiments, bias, fraud and data boundaries
Expose interference, selection bias and false positives before interpreting a metric.
- Liu Peng & Wang Chao · Computational Advertising, 2nd ed. Public contents: related technology; bias/interference exercise by the Lab
Boundary: Constructed mixture bias, budget interference and false positives; not real fraud accuracy or an online experiment.
A18
End-to-end ledger and synthesis
Reconcile eligibility, decision, feedback and billing with idempotent events.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial PDF pages 18–19: systems background; ledger designed by the Lab
Boundary: A single-request, single-process idempotency mechanism; no concurrent transactions, refunds, real clicks or revenue.
Original lecture library 2 entries
These links open the original author or publisher material, in its original language and under its own terms.
- Introduction to Computational Communication · Chapter 8
96 slides: products 14–41; market mechanisms 42–94 · Chinese
A pinned 96-slide Chinese chapter on products, objectives, games, matching and auctions. Not a complete advertising engineering course.
- Weinan Zhang / Jian Xu · CIKM 2016 RTB Tutorial
164-page PDF: unit links open the relevant page · English
164 author slides covering response prediction, bidding, attribution, pacing, targeting and reserve prices. Lab mechanisms are independently authored, not full paper replications.