A Game Theory Approach for Negotiating in Data Marketplaces

We introduce a game-theoretic framework, together with a distributed algorithm, for automating multi-issue data-sharing negotiations. The framework models interactions between data providers and data consumers as a bargaining problem and defines a number of bargaining actions, negotiation parameter categories (such as numerical, hierarchical, or query-based) and associated classes of utility functions for negotiating on these parameters. We instantiate the framework for data-sharing in data marketplaces. For this real-world use case, we provide a set of concrete negotiation parameters such as properties of the dataset (e.g., format, price, or date), views or query-defined subsets, and even governance-related ones such as negotiating on the privacy policy or the terms-of-use of the dataset. We propose two heuristics to guide agents' behaviours, greedy and concession-based. For our framework and heuristics, we provide theoretical guarantees as well as an implementation evaluated on real-world datasets from bank marketing, gym workouts, and sensing applications. Compared to traditional negotiation approaches, such as monotonic concessions and take-it- or-leave-it protocols, our approach achieves higher agreement rates and utilities, while up to 49% improves utility balance, measured as fairness.