Towards Causality-centric Recommender Systems
- Publication Type:
- Thesis
- Issue Date:
- 2025
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Recommender Systems (RecSys) have become pivotal tools in tackling data overload, for information provision in applications across academia and industry. Most RecSys innovations rely on correlation-based learning, such as collaborative filtering, to model behavioral patterns and predict user preferences. However, user behavior is a mixed reflection of user interests, driven by various factors. Thus, it is crucial to go beyond correlation learning to model the true user interest. This thesis aims to incorporate causal learning theory into recommendations. Unlike correlation-based methods, causal learning identifies cause-effect relationships - where a cause (e.g., user interest) leads to an effect (e.g., user behavior). Causal relationships provide a stronger foundation for recommendation models and can be leveraged in various applications. This thesis addresses key challenges in RecSys, including data bias, model accuracy, explainability, and fairness, by designing causality-driven solutions tailored to each problem. Through extensive experiments, these causality-centric methods demonstrate clear advantages over traditional correlation-based approaches, achieving higher accuracy, improved explainability, enhanced fairness, and greater generalization ability.
Please use this identifier to cite or link to this item: