Unsupervised/Semi-supervised Data Valuation based on Coopeative Games

Abstract

With the increasing importance of data analytics, it has become important to properly value data and fairly reward data providers. Based on the concept of the Shapley value in cooperative game theory, methods have been proposed to evaluate the value of data according to its contribution to improving the accuracy of machine learning models. These methods are based on the assumption that the data collector has sufficient test data and can accurately evaluate the accuracy of the models. However, in practice, data collectors often have no or little test data in advance. In this paper, we formulate the problem of data value evaluation in such cases as unsupervised/semi-supervised data valuation and discuss their solutions.