机器学习模型交易中的数据购买量与模型定价  

Data Purchase Volume and Model Pricing in Machine Learning Model Transactions

作　　者：林苗君 羊梓敏 陈斌 LIN Miaojun;YANG Zimin;CHEN Bin(Financial Department,Huaqiao University,Quanzhou 362021,China;School of Mathematical Science,Huaqiao University,Quanzhou 362021,China)

机构地区：[1]华侨大学财务处,福建泉州362021 [2]华侨大学数学科学学院,福建泉州362021

出　　处：《华侨大学学报(自然科学版)》2025年第1期95-103,共9页Journal of Huaqiao University(Natural Science)

基　　金：国家自然科学基金资助项目(12071165);福建省自然科学基金资助项目(2023J01124);中央高校基本科研业务费专项资金资助(ZQN-1102)。

摘　　要：基于数据边界的Shapley值总和构建成本分配问题,采用截断蒙特卡洛的快速算法计算Shapley值,证明了数据边界最优解存在性。针对不同版本模型定价场景,定义模型经纪人收入最大化模型的定价问题,并将收入最大化模型的定价问题转换为等价整数线性规划问题。运用公共数据集数值验证文中方法的正确性,同时与已有的4种方法进行实验对比。实验结果表明:文中方法可以提高模型经纪人收入和模型买方的购买比例。Cost allocation problem based on Shapley value summation on data boundaries is constructed.Using the fast algrithm of the Truncated Monte Carlo,the existence of the optimal solution of the data boundary is proved.Aiming at pricing scenarios of different versions of the models,the pricing problem of the income maximizing model of the model broker is defined,and the pricing problem of the income maximizing model is transformed into an equivalent integer linear programming problem.By public datasets,our proposed method is validated its correctness,and the experiment is compared with four existing methods,simultaneously.The experimental results show that the proposed method can increase the income of the model broker and the purchase ratio of the model buyer.

关 键 词：数据定价 模型定价 SHAPLEY值 整数规划 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置] F49[自动化与计算机技术—控制科学与工程]