Learning Individual Treatment Effects under Heterogeneous Interference in Networks.
C. Zhao,Yuqi Bai,Ruoxuan Xiong,Qingyu Cao,Chao Ma,Ning Jiang,Fei Wu,Kun Kuang
DOI: https://doi.org/10.1145/3673761
IF: 4.157
2024-01-01
ACM Transactions on Knowledge Discovery from Data
Abstract:Estimates of individual treatment effects from networked observational dataare attracting increasing attention these days. One major challenge in networkscenarios is the violation of the stable unit treatment value assumption(SUTVA), which assumes that the treatment assignment of a unit does notinfluence others' outcomes. In network data, due to interference, the outcomeof a unit is influenced not only by its treatment (i.e., direct effects) butalso by others' treatments (i.e., spillover effects). Furthermore, theinfluences from other units are always heterogeneous (e.g., friends withsimilar interests affect a person differently than friends with differentinterests). In this paper, we focus on the problem of estimating individualtreatment effects (both direct and spillover effects) under heterogeneousinterference. To address this issue, we propose a novel Dual WeightingRegression (DWR) algorithm by simultaneously learning attention weights thatcapture the heterogeneous interference and sample weights to eliminate thecomplex confounding bias in networks. We formulate the entire learning processas a bi-level optimization problem. In theory, we present generalization errorbounds for individual treatment effect estimation. Extensive experiments onfour benchmark datasets demonstrate that the proposed DWR algorithm outperformsstate-of-the-art methods for estimating individual treatment effects underheterogeneous interference.