PatentXAI framework uses Shapley values to price patents inside complex products

Patents are notoriously hard to price individually, especially when a single product embodies tens of thousands of them. Joy Bose's new arXiv preprint introduces PatentXAI, a framework that treats patent valuation as an explainable AI problem. It uses Shapley values—a cooperative game theory concept that fairly distributes profit based on each patent's marginal contribution—to allocate revenue across subsets of patents. To make computation tractable, the method restricts each patent's coalition to its Markov Blanket inside a knowledge graph, grounded in the C-SVE conditional independence theorem. Scaling experiments from n=12 to n=100 patents show a median Markov Blanket size of 32.9% of n at n=100 (90th percentile: 55.2%), and a runtime of 10 milliseconds per patent. Accuracy differences against exact ground truth at n=12 is 0.088; against a high-sample Monte Carlo reference at n=100 is 0.062±0.003. When 80% of patents share a dense component, the blanket expands to cover that cluster, improving accuracy to 0.039 difference vs. reference. Profit allocation proceeds hierarchically: exact Shapley distributes total profit among macro-components, then centrality-weighted Shapley distributes each component budget among covering patents.

The framework explicitly identifies the main open problem as estimating the characteristic function v(S)—the revenue achievable by any patent subset—from real data. Bose distinguishes this from the computational contribution and outlines a roadmap for empirical validation using public ETSI, USPTO, and this http URL datasets. For tech professionals, this could change how R&D investments are justified, patent portfolios are valued, and licensing fees are determined. While the paper is theoretical, its practical implications for intellectual property economics are sizable.