AIE: Auction Information Enhanced Framework for CTR Prediction in Online Advertising

TL;DR

This paper tackles CTR prediction in online advertising under a highly dynamic auction environment, where traditional models suffer from auction-induced data bias. It introduces Auction Information Enhanced Framework (AIE), comprising two training-time modules: Adaptive Market-price Auxiliary Module (AM2) and Bid Calibration Module (BCM), designed to exploit posterior auction signals without increasing inference latency. AM2 learns a fine-grained market-price auxiliary task via a scenario-conditioned dynamic network, while BCM reweights high-bid positive samples to calibrate the training distribution toward the online target. Across public and industrial datasets, plus a one-month online A/B test, AIE yields consistent gains in AUC and revenue-oriented metrics (eCPM, Rev, RevNDCG) and demonstrates strong compatibility with diverse backbones, reducing predicted bias and improving practical impact.

Abstract

Click-Through Rate (CTR) prediction is a fundamental technique for online advertising recommendation and the complex online competitive auction process also brings many difficulties to CTR optimization. Recent studies have shown that introducing posterior auction information contributes to the performance of CTR prediction. However, existing work doesn't fully capitalize on the benefits of auction information and overlooks the data bias brought by the auction, leading to biased and suboptimal results. To address these limitations, we propose Auction Information Enhanced Framework (AIE) for CTR prediction in online advertising, which delves into the problem of insufficient utilization of auction signals and first reveals the auction bias. Specifically, AIE introduces two pluggable modules, namely Adaptive Market-price Auxiliary Module (AM2) and Bid Calibration Module (BCM), which work collaboratively to excavate the posterior auction signals better and enhance the performance of CTR prediction. Furthermore, the two proposed modules are lightweight, model-agnostic, and friendly to inference latency. Extensive experiments are conducted on a public dataset and an industrial dataset to demonstrate the effectiveness and compatibility of AIE. Besides, a one-month online A/B test in a large-scale advertising platform shows that AIE improves the base model by 5.76% and 2.44% in terms of eCPM and CTR, respectively.

Figures (4)

• Figure 1: (a) CTRs against different market prices under four industries in an industrial dataset. (b) Market price distribution under four industries in an industrial dataset.
• Figure 2: Auction Bias Illustration. When the advertiser gives a normal bid for an ad as Figure (a) shows, the ad only wins the auction in Context 1 due to the high click relevance. When the advertiser gives a high bid for the same ad as Figure (b) shows, the ad wins more auctions in all three Contexts though the click relevance is low for Context 2 and 3. Therefore, the training data distribution is different from the original target distribution, leading to data bias, which is called auction bias due to high bidding.
• Figure 3: Overall framework of AIE, which consists of two key modules: AM2 and BCM. AM2 uses market price and scenario features to construct an auxiliary task. BCM uses bid to impact the cross-entropy loss.
• Figure 4: Ablation study about different modules of AIE in terms of four metrics.