Causal Inference for Recommender Systems: Methods, Challenges, and Applications

Abstract

Traditional recommendation systems based on user interaction data for correlation analysis have significant bias issues—such as selection bias, exposure bias, and popularity bias —that arise from the inherent limitations of observational data, making it difficult to predict user preferences accurately without bias and often leading to homogenized recommendations that fail to uncover users’ potential interests. Users lose trust in recommendation systems because of this situation which makes it difficult to implement these systems in complex environments that need exact personalized healthcare services and intelligent decision systems. The review examines causal recommendation surveys and presents new approaches (debiasing and causal collaborative filtering and distillation) and demonstrates their applications in time-series analysis and healthcare and uplift modeling. The research demonstrates that causal inference methods remove confounding variables which produces improved prediction accuracy and better system operation understanding for users. Research on causal inference will progress through the combination of deep learning techniques which analyze intricate temporal relationships and multiple domains to create functional recommendation platforms. The research establishes vital foundations for upcoming investigations which will link causal inference to recommendation systems through its complete analysis of this field.