A Framework for Graph-Conditioned Hierarchical Shapley Attribution in Patent Valuation

Shapley values offer a mathematically fair way to assign patent value: The framework uses Shapley values—a game theory concept that satisfies efficiency, symmetry, and additivity principles—to determine each patent's "fair share" of product profit, addressing a long-standing problem in intellectual property economics where thousands of patents contribute to a single product.

Graph-based shortcuts make fair valuation computationally practical: By restricting each patent's contribution analysis to its "Markov Blanket" (related patents in a knowledge graph) rather than all possible combinations, the method reduces computation time to 10 milliseconds per patent while maintaining accuracy within 0.062 of reference values, making fair attribution feasible at scale.

The approach becomes more accurate when patents naturally cluster together: When 80% of patents share a common component, the method's accuracy improves (difference drops to 0.039), suggesting that the framework better handles realistic patent portfolios where related technologies are interdependent rather than independent.

Hierarchical allocation enables transparent profit distribution: The framework distributes total product profit first among major technology components using exact Shapley values, then allocates within components using centrality measures, creating a transparent, multi-level explanation of how value flows from individual patents to overall product revenue.