Learning Individual Treatment Effects under Heterogeneous Interference in Networks

• DOI: 10.1145/3673761
• Publication Date: 2024-06-19T06:38:54Z
• Authors: Ziyu Zhao; Yuqi Bai; Ruoxuan Xiong; Qingyu Cao; Chao Ma; Ning Jiang; Fei Wu; Kun Kuang
• Keywords: Benchmark (surveying); Spillover effect; Sample (material)

Estimating individual treatment effects in networked observational data is a crucial and increasingly recognized problem. One major challenge of this problem violating the stable unit value assumption (SUTVA), which posits that unit’s outcome independent others’ assignments. However, network data, influenced not only by its (i.e., direct effect) but also treatments others spillover since presence interference. Moreover, interference from other units always heterogeneous (e.g., friends with similar interests have different influence than those interests). In article, we focus on estimating (including effect under networks. To address problem, propose novel dual weighting regression (DWR) algorithm simultaneously learning attention weights to capture neighbors sample eliminate complex confounding bias We formulate process as bi-level optimization Theoretically, give generalization error bound for expected estimation effects. Extensive experiments four benchmark datasets demonstrate proposed DWR outperforms state-of-the-art methods