Data Pricing in Vertical Federated Learning

Federated Learning (FL) creates an ecosystem for multiple parties to collaborate on building models while protecting data privacy for the participants. As a critical category of FL, Vertical federated learning (VFL) is mainly used to model heterogeneous data from multiple parties. In order to scientifically and fairly distribute the revenue of federated participants in VFL, this paper provides a scientific and fair-minded data pricing method based on the contribution of participants for federated models. Firstly, a fair and accurate measurement method of the contribution of each federated participant is provided based on shapely values. On this basis, a data pricing model based on Stackelberg with the hosts as the leader and the guest as the follower is formulated in VFL. The numerical solutions for the data-pricing model indicate that it outperforms traditional data pricing methods such as query-based fixed pricing. These results also provide managerial guidance on contribution measurement, data pricing, and revenue distribution for FL platform owners.