Assisting the Social Scientist through Human-Machine Consensus: Causal Discovery Approach to Unravel Cyberbullying

Cyberbullying is a pressing issue among minors that demands thoughtful intervention strategies. While causal inference has become popular in many scientific domains, we argue that the goal in social sciences should not be to automate causal discovery, but to support expert-driven understanding through simple, explainable models. We present a framework that builds probabilistic graphical causal models using a consensus-based process between domain experts and structure learning algorithms implemented in GeNIe (PC, Bayesian Search, and GTT). These tools allow easy incorporation of expert constraints, which is essential given the limited sample sizes and measurement challenges typical in social sciences. Rather than emphasizing the discovery of novel causal structures, we show how combining expert insights with machine learning improves model validity, interpretability, and usefulness. Our method provides a principled way to explore interventional questions while acknowledging the strengths and limits of both human and automated reasoning.