Causal Inference

Causal inference seeks to identify why events occur—identifying causal relationships between variables rather than mere correlations. In the context of fake news, causal inference addresses questions like "what causes users to share fake news?" rather than "who is correlated with sharing fake news?"

Observational data (e.g., social media logs) presents a fundamental challenge: we can only observe one outcome per user, not the counterfactual ("what would this user have done under a different condition?"). Confounders—variables that influence both the treatment (e.g., user attribute) and outcome (e.g., sharing behavior)—create spurious associations if left uncontrolled.

Key papers

• Causal Machine Learning: A Survey and Open Problems — comprehensive survey of causal ML methods across supervised learning, generative modeling, explanations, fairness, and RL with taxonomy of open problems
• Causal Understanding of Fake News Dissemination on Social Media — applies inverse propensity scoring to learn unbiased user embeddings as confounders

Related topics

• Selection Bias (data collection bias that violates causal assumptions)
• Confounding (unmeasured variables affecting treatment and outcome)
• User Attributes (profile-based treatments in causal models)