Learn then Decide: A Learning Approach for Designing Data Marketplaces
Yingqi Gao, Jin Zhou, Hua Zhou, Yong Chen, Xiaowu Dai · 2025
Read Learn then Decide: A Learning Approach for Designing Data Marketplaces by Yingqi Gao, Jin Zhou, Hua Zhou, Yong Chen, Xiaowu Dai (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-10773.txt.
Introduction
Yingqi Gao and co-authors design a better way to price data in online marketplaces. Sellers want strong revenue, while buyers want fair prices that do not punish some users over others. Their mechanism, called MAPP, learns first and prices second.
In stage one, auctions reveal how much bidders value the data, building an estimate of the value distribution. In stage two, that estimate sets a single posted price. The design keeps bidding truthful and each buyer willing to take part. Theory ties pricing regret to how well the value curve is estimated, and the method keeps that regret small.
Catalog record
| CREATOR | Yingqi Gao, Jin Zhou, Hua Zhou, Yong Chen, Xiaowu Dai |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-10773
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Learn then Decide: A Learning Approach for Designing Data Marketplaces Yingqi Gao, Jin Zhou, Hua Zhou, Yong Chen, Xiaowu Dai Abstract As data marketplaces become increasingly central to the digital economy, it is crucial to design efficient pricing mechanisms that optimize revenue while ensuring fair and adaptive pricing. We introduce the Maximum Auction-to-Posted Price (MAPP) mechanism, a novel two-stage approach that first estimates the bidders’ value distribution through auctions and then determines the optimal posted price based on the learned distribution. We establish that MAPP is individually rational and incentive-compatible, ensuring truthful bidding while balancing revenue maximization with minimal price discrimination. MAPP achieves a regret of O p ( n − 1 ) O_{p}(n^{-1}) when incorporating historical bid data, where n n is the number of bids in the current round. It outperforms existing methods while imposing weaker distributional assumptions. For sequential dataset sales over T T rounds, we propose an online MAPP mechanism that dynamically adjusts pricing across datasets with varying value distributions. Our approach achieves no-regret learning, with the average cumulative regret converging at a rate of O p ( T − 1 / 2 ( log T ) 2 ) O_{p}(T^{-1/2}(\log T)^{2}) . We validate the effectiveness of MAPP through simulations and real-world data from the FCC AWS-3 spectrum auction. Key Words : Auctions; Data marketplace; Nonparametric density estimation; Online learning; Revenue maximization. 1 Introduction The economic value of data has grown significantly in recent decades, driving its role as a key resource for artificial intelligence and decision making ( Veldkamp and Chung 2024 ) . The increasing demand for data has spurred the rise of data marketplaces such as Amazon AWS Data Exchange, Datarade, and Snowflake Data Marketplace. However, these platforms remain inefficient, as transactions often involve prolonged price negotiations, resulting in high transaction costs ( Kennedy et al. 2022 ) . This inefficiency raises the question: Can alternative pricing mechanisms be designed to not only maximize seller revenue but also enhance the efficiency of data trading? From the seller’s perspective, an effective solution should not only reduce transaction costs but also maximize revenue. Auction theory naturally serves this purpose, with classic results by Myerson 1981 demonstrating that auctions consistently outperform negotiations ( Bulow and Klemperer 1996 ) . Reserve prices play a central role in this framework ( Kanoria and Nazerzadeh 2021 ) , but when the value distribution is unknown, setting effective reserve prices becomes challenging ( Chawla et al. 2014 ) . Recent empirical evidence suggests that auctions are losing favor in real-world platforms such as eBay, where sellers increasingly opt for posted prices due to intensified competition and buyers’ preference for convenience ( Einav et al. 2018 ) . This shift highlights the need for hybrid mechanisms that combine the revenue efficiency of auctions with the accessibility of posted pricing. This paper introduces the Maximum Auction-to-Posted Price (MAPP) mechanism, a novel two-stage pricing strategy aimed at optimizing revenue in data marketplaces while minimizing price discrimination. The mechanism operates in two phases. In the auction phase, a fixed number of buyers submit bids, and each buyer is assigned a bid-independent price determined by the bids of other participants rather than their own. This approach ensures incentive compatibility, as buyers are motivated to bid truthfully. Buyers must purchase the dataset if their bid meets or exceeds their assigned price. The posted-price phase follows, during which the seller sets the highest observed auction price as the posted price. Subsequent buyers who arrive dynamically decide to purchase only if their valuations meet or exceed this price. The workflow of MAPP is illustrated in Figure 1 . Figure 1 : Illustration of an online algorithm based on the MAPP mechanism, where T is the total number of rounds. MAPP improves upon existing methods for selling goods with unlimited supply by combining auction-driven price discovery ( Segal 2003 ; Goldberg et al. 2006 ) to optimize revenue from early buyers with a posted-price mechanism ( Bar-Yossef et al. 2002 ; Kleinberg and Leighton 2003 ; Broder and Rusmevichientong 2012 ) to ensure accessibility for later arrivals. Unlike existing approaches, MAPP fixes the posted price at the highest observed auction price, effectively reducing price discrimination while preserving revenue efficiency. Additionally, MAPP satisfies individual rationality, ensuring voluntary participation, a key economic property that drives market efficiency. The effectiveness of MAPP relies on accurately estimating the value distribution to set a near-optimal price. If the distribution were known, the optimal price would be clear ( Myerson 1981 ) . However, in practice, uncertainty about the value distribution leads to regret, which quantifies the resulting revenue loss. Aligned with Segal 2003 , we began with the commonly used kernel density estimation (KDE) approach. However, we derive an improved regret bound with finite-sample guarantees, relax its underlying assumptions, and mitigate the price discrimination inherent in the mechanism of Segal 2003 . When historical value observations from datasets with distributions in the same family (e.g., Gaussian) are available, we further improve the regret bound using the repeated density estimation (RDE) method of ( Qiu et al. 2022 ) . This approach achieves a regret rate of O_{p}(n^{-1}) rate, a significant improvement over the existing O(n^{-2/3}(\log n)^{3/2}) rate ( Segal 2003 ; Kleinberg and Leighton 2003 ; Broder and Rusmevichientong 2012 ) , where n denotes the number of bids received for the current dataset. RDE enhances value distribution estimation by applying functional principal component analysis to log-ratio transformed kernel-based estimates derived from historical observations of similar datasets. This technique uncovers a latent low-dimensional structure in the shared distribution family, effectively reducing the feasible space for nonparametric density estimation. Observing that real-world data sellers often own multiple datasets from the same or similar sources, typically sharing a common value distribution family with varying parameter values, such as Gaussian distributions with different means and variances, we extend MAPP to the online setting for selling these datasets. To the best of our knowledge, this is the first online algorithm designed to price unlimited-supply goods with varying value distributions across different goods. The online MAPP mechanism consists of two stages: an exploration stage, where value distributions are estimated using KDE to gather market insights, and an exploitation stage, where pricing strategies for subsequent datasets are refined using RDE, leveraging bid data from the exploration stage for more accurate distribution estimation. We establish that the online MAPP mechanism achieves an average cumulative regret of O_{p}(T^{-1/2}(\log T)^{2}) , where T denotes the total number of rounds or datasets. This ensures no-regret learning, with average regret diminishing as the number of sales increases. Empirical evaluations, including simulations and real-world auction data from the FCC AWS-3 spectrum auction, validate the convergence of the average cumulative regret, demonstrating the effectiveness of MAPP in data marketplaces. The rest of this paper is structured as follows: Section 2 provides background on data marketplaces. Section 3 introduces the MAPP mechanism, establishes its incentive guarantees, and extends it to the online setting. Section 4 presents the regret analysis on both instantaneous and cumulative regrets. Section 5 evaluates the mechanism through empirical studies using simulated and real-world data. Section 6 reviews related literature. Section 7 concludes the paper with future directions. 2 Background 2.1 Model Setup We