With the advancement of internet technologies, daily life has become more convenient; however, cyberbullying victimization has also risen. It poses serious threats to psychological and physical well-being and is a potential risk factor for suicide. Although prior research has linked cyberbullying victimization to suicide risk, evidence on the variation of this association across individuals and its robustness to high-dimensional confounders remains limited. This study estimates the average treatment effect (ATE) and conditional average treatment effect (CATE) of cyberbullying victimization on suicide risk.

Using 6,476 Weibo users’ posts (3,873 individuals in the cyberbullying victimization group and 2,603 individuals in the control group), we extracted linguistic and psychological features via dictionary-based methods and predictive models, and applied Double Machine Learning (DML) to estimate high-dimensional adjusted associations between cyberbullying victimization and suicide-related language expression. SHapley Additive exPlanations (SHAP) and tree-based explainers quantified the relative importance of features and identified subgroups with distinct CATEs.

DML analyses indicated a positive DML-adjusted contrast between the victimization and non-victimization groups on the CSD-derived suicide-related language score (ATE = 0.34, 95% CI = [0.20, 0.48]), with heterogeneous effects across psychological and linguistic traits. Hopelessness plays a central role, while elevated Hostility and Tentative further increase risk; greater Cause, Autonomy, and Money are associated with attenuated effects.

By integrating DML, this study provides preliminary evidence regarding exploratory heterogeneity patterns in the DML-adjusted association between cyberbullying victimization and CSD-derived suicide-related language score and offers a theoretical basis for identifying high-risk victims and designing targeted interventions.