关键词: 数据价值, 数据估值, 因果推断, 合成控制

Abstract: To promote the development of data trading and markets by addressing the challenges of high cost andcomputational complexity in traditional data valuation methods, this paper proposes a novel causal inference-basedmethod for assessing the incremental value of data from a buyer's perspective: the Data Synthetic Control Method(DSCM). Viewing the addition of new data as an intervention, DSCM innovatively constructs a counterfactualinference framework to precisely quantify the contribution of new data to the performance of supervised machinelearning models. In simulated experiments, the value estimation from DSCM method shows high concordance withactual observed values for various machine learning models, with an average error of only 0.0032. In a real-worldapplication of advertising click-through rate prediction, the value estimation from DSCM method of user-centricdata exhibits an average error rate of only 9% compared to the actual performance lift, significantly outperformingtraditional valuation methods. These evidences show that the proposed method can provide accurate and stable datavalue assessment and effectively support enterprises' data-driven decision-making.

Key words: data value, data evaluation, causal inference, synthetic control