Intelligent Advertising Algorithm System at iQIYI: Architecture, Smart Bidding, Inventory Allocation, and TrueView

Advertising is a key monetization channel for internet traffic and a major driver for large‑scale AI applications in industry. iQIYI's advertising business has grown to a hundred‑billion‑yuan scale, covering both brand ads (pre‑roll video) and performance ads (in‑feed). The company built an intelligent advertising algorithm system to support these scenarios.

System Framework : The system integrates smart bidding, inventory allocation, and TrueView modules, leveraging AI to improve advertiser efficiency, platform revenue, and user experience.

Smart Bidding : Performance ads use eCPM (eCPM = bid × pCTR × pCVR) for dynamic pricing. A machine‑learning based bidding algorithm is employed: first, a K‑Nearest‑Neighbors (KNN) approach finds similar historical orders; if similar orders exist, their data are used to fit the bid‑exposure relationship and suggest a bid; otherwise, an XGBoost model explores bid suggestions from a broader sample space.

Inventory Allocation : Brand ads are sold via contracts, making efficient inventory allocation critical. The problem is NP‑hard, so a hybrid offline‑online solution is designed. Offline, ARIMA and LSTM models forecast future inventory across multiple dimensions. The offline forecast constructs a bipartite allocation graph for pre‑allocation. Online, real‑time order progress adjusts the offline predictions to achieve global optimality.

TrueView : TrueView is a video ad format where users voluntarily watch ads; advertisers pay only for views longer than 30 seconds. To maximize TrueView rates, iQIYI built a user‑segmentation based TrueView prediction algorithm. Traditional models (Logistic Regression, GBDT, FM) and a Wide&Deep deep model were evaluated, with the latter delivering notable performance gains.

Future Outlook : AI will continue to make advertising more intelligent, improving matching between ads and consumers. As NLP, image, and video understanding technologies advance, they will further enhance user and content comprehension, strengthening the relevance of ads to users and video content.