Online Competitive Posted-Pricing Mechanism for Trading Time-Sensitive Valued Data

With the rapid growth of the applications in AI, the demand of data has increased significantly. To facilitate the circulation of data, numerous online data sharing and trading platforms (aka, data brokers) have emerged. In many situations the data value is time-sensitive, depending on the freshness of the data. Such a time-dependence can be characterized by a discount function $d(t)$ representing the data value fluctuation factor over the lifetime $t$ . In this work, we propose a series of posted-price mechanisms to for data trading. By assuming the buyers' initial valuations follow a given distribution, we design two online posted-price mechanisms which are constant-competitive (approximately maximize the data seller's revenue), value truthful and (semi)-time truthful. We then extend our mechanisms to general value distributions. Finally, we propose an online learning mechanism to resolve the issue that we do not know the parameters of the distribution, and further explore the method of guaranteeing truthfulness. Our simulation results show that our mechanisms achieve more than 90% of the revenue over the offline baseline mechanism, and the competitive ratios converge to 1 as the number of buyers increases.