Knowledge-informed Bidding with Dual-process Control for Online Advertising

Huixiang Luo; Longyu Gao; Yaqi Liu; Qianqian Chen; Pingchun Huang; Tianning Li

cs.AIMar 5, 2026

Knowledge-informed Bidding with Dual-process Control for Online Advertising

Authors: Huixiang Luo, Longyu Gao, Yaqi Liu, Qianqian Chen, Pingchun Huang, Tianning Li

Organizations: Alibaba Health, Hangzhou, China · Alibaba Health, Beijing, China

Abstract

Bid optimization in online advertising relies on black-box machine-learning models that learn bidding decisions from historical data. However, these approaches fail to replicate human experts' adaptive, experience-driven, and globally coherent decisions. Specifically, they generalize poorly in data-sparse cases because of missing structured knowledge, make short-sighted sequential decisions that ignore long-term interdependencies, and struggle to adapt in out-of-distribution scenarios where human experts succeed. To address this, we propose KBD (Knowledge-informed Bidding with Dual-process control), a novel method for bid optimization. KBD embeds human expertise as inductive biases through the informed machine-learning paradigm, uses Decision Transformer (DT) to globally optimize multi-step bidding sequences, and implements dual-process control by combining a fast rule-based PID (System 1) with DT (System 2). Extensive experiments highlight KBD's advantage over existing methods and underscore the benefit of grounding bid optimization in human expertise and dual-process control.

Figures & tables

Figure 1: The architecture of the KBD method. Blue-shaded components denote black-box modules implemented via neural networks, yellow-shaded components indicate rule-based interpretable modules, and mixed-color regions represent their fusion.

Method	$R/R^{*}\uparrow$	constraint satisfaction $\uparrow$
ARTEO ( Korkmaz et al., 2022 )	0.601	80.65%
BOCO ( Liang et al., 2023 )	0.618	74.10%
SPB ( Su et al., 2024b )	0.699	75.32%
PUROS ( Luo et al., 2024 )	0.723	81.73%
GCB-safe ( Castiglioni et al., 2025 )	0.606	80.77%
KBD w/o DT	0.706	81.43%

Table 1: Results of click maximization on iPinYou dataset.

Experimental Period	Group	Group Detail			cost-exhaust ratio	GMV	Campaign Duration
Experimental Period	Group	use PID	use IEFormer	use DT	cost-exhaust ratio	GMV	Campaign Duration
2025/09/28-2025/10/18	Control Treatment	✓ ✓	✓		/ + 8.44%	/ + 6.14%	/ + 0.58%
2025/11/20-2025/12/10	Control Treatment	✓ ✓	✓	✓	/ +14.55%	/ +13.01%	/ +1.36%
2025/12/11-2025/12/31	Control Treatment	✓ ✓	✓ $\checkmark$	✓	/ +2.27%	/ +2.66%	/ +0.20%

Table 2: Online test results on ECA : relative lifts against control group; absolute metrics redacted for commercial confidentiality.

Figure 2: Case studies on the influence of

L_{\mathrm{margin}}

. As highlighted by the ellipsoidal regions, the price-volume curve is more prone to overfitting the noisy data without

L_{\mathrm{margin}}

Method	wmape $\downarrow$	mape $\downarrow$	perf10 $\uparrow$
w/o IE	0.2588	0.2348	0.2738
w/o GLA	0.2412	0.2270	0.2793
w/o $L_{\mathrm{margin}}$	0.2386	0.2268	0.2921
w/o tROI data	0.2226	0.2600	0.2976
KBD	0.2187	0.2151	0.3214

Table 3: Ablation Study of IML on ECA dataset.

Method	wmape $\downarrow$	mape $\downarrow$
IGCM ( Tang et al., 2025 )	0.9979	0.9949
GP ( Schulz et al., 2018 )	0.6414	1.8021
CMNN ( Li et al., 2024 )	0.6309	2.0599
DIPN ( Shen et al., 2021 )	0.6308	1.7647
SMNN ( Kim and Lee, 2024 )	0.6179	2.2503
SMM ( Igel, 2023 )	0.6023	1.6785

Table 4: Price-volume model performance on ECA dataset.

Figure 3: Robustness of IEFormer to segment number N.

Automated Bidding

Automated bidding studies algorithms that choose auction bids over time while balancing value, budgets, and performance constraints.

See all 36 papers

JD-BP: A Joint-Decision Generative Framework for Auto-Bidding and Pricing
Apr 7, 2026Linghui Meng, Chun Gan, Shengsheng Niu +8Automated Bidding Real-Time Bidding
JD.com China · The Chinese University of Hong Kong China
DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation
Jun 12, 2026Miduo Cui, Haochen Wang, Shangqin Mao +6Distributional RL Automated Bidding
Shandong University, Jinan, China · Meituan, Beijing, China
Beyond Single-Episode Optimization: Sliding-Window Aware Generative Auto-Bidding for Long-Term Advertising Effectiveness
Jul 28, 2026Binglin Wu, Chuan Yue, Yingyi Zhang +4Automated Bidding Online Advertising
Dalian University of Technology · Alibaba International Digital Commerce Group · Dalian University of Technology & City University of Hong Kong

Apr 7, 2026J/K move · Enter open · S save

JD-BP: A Joint-Decision Generative Framework for Auto-Bidding and Pricing

Linghui Meng, Chun Gan, Shengsheng Niu, Chengcheng Zhang+7

JD.com China · The Chinese University of Hong Kong China

Automated Bidding Real-Time Bidding Online Advertising Dynamic Pricing

Open paper AskPDF arXiv

Auto-bidding services optimize real-time bidding strategies for advertisers under key performance indicator (KPI) constraints such as target return on investment and budget. However, uncertainties such as model prediction errors and feedback latency can cause bidding strategies to deviate from ex-post optimality, leading to inefficient allocation. To address this issue, we propose JD-BP, a Joint generative Decision framework for Bidding and Pricing. Unlike prior methods, JD-BP jointly outputs a bid value and a pricing correction term that acts additively with the payment rule such as GSP. To mitigate adverse effects of historical constraint violations, we design a memory-less Return-to-Go that encourages future value maximizing of bidding actions while the cumulated bias is handled by the pricing correction. Moreover, a trajectory augmentation algorithm is proposed to generate joint bidding-pricing trajectories from a (possibly arbitrary) base bidding policy, enabling efficient plug-and-play deployment of our algorithm from existing RL/generative bidding models. Finally, we employ an Energy-Based Direct Preference Optimization method in conjunction with a cross-attention module to enhance the joint learning performance of bidding and pricing correction. Offline experiments on the AuctionNet dataset demonstrate that JD-BP achieves state-of-the-art performance. Online A/B tests at JD.com confirm its practical effectiveness, showing a 4.70% increase in ad revenue and a 6.48% improvement in target cost.

Abstract

Automated Bidding

Explore similar work

JD-BP: A Joint-Decision Generative Framework for Auto-Bidding and Pricing

DRIVE: Distributional and Retrieval-Augmented Bidding with Value Evaluation

Beyond Single-Episode Optimization: Sliding-Window Aware Generative Auto-Bidding for Long-Term Advertising Effectiveness