A survey on causal inference for recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Causal inference is increasingly important in recommender systems (RS) for understanding cause-and-effect.
• Existing surveys often classify methods by practical issues, fragmenting unified causal theories.
• RS researchers need a coherent, theory-driven perspective on causal inference.

Purpose of the Study:

• To systematically review causal inference in RS from a causal theory standpoint.
• To propose a novel taxonomy for categorizing existing causal inference methods in RS.
• To facilitate deeper integration of causal inference within the RS field.

Main Methods:

• Introduces fundamental concepts of causal inference.
• Proposes a theory-driven taxonomy: potential outcome framework, structural causal model, and general counterfactuals.
• Reviews technical details of applying causal inference to RS problems.

Main Results:

• Provides a systematic review of causal inference in RS literature.
• Organizes existing methods into a novel taxonomy based on underlying causal theories.
• Identifies key challenges and future research directions in causal recommender systems.

Conclusions:

• A theory-driven approach is crucial for advancing causal inference in RS.
• The proposed taxonomy offers a unified perspective on diverse causal methods.
• Further research is needed to fully leverage causal inference for improved recommender systems.