Disentangled Representation Learning for Recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• User behavior in recommender systems is driven by complex, entangled latent factors.
• Existing methods struggle to disentangle these factors, limiting interpretability and control.
• Uncovering disentangled latent factors is crucial for robust representation learning.

Purpose of the Study:

• To develop a novel model for learning disentangled representations from user behaviors.
• To address the challenges posed by entangled latent factors in recommender systems.
• To enhance the interpretability and controllability of recommender systems.

Main Methods:

• Proposed the SEMantic MACRo-mIcro Disentangled Variational Auto-Encoder (SEM-MacridVAE) model.
• Achieved macro disentanglement via prototype routing for high-level user intentions.
• Ensured micro disentanglement using an information-theoretic regularizer for low-level preferences.
• Incorporated item semantic information (visual, categorical) to boost performance.

Main Results:

• SEM-MacridVAE achieved significant improvements over state-of-the-art baselines.
• Demonstrated the interpretability and controllability of the learned representations.
• Empirical experiments validated the model's effectiveness.

Conclusions:

• The SEM-MacridVAE model offers a new approach to disentangled representation learning in recommender systems.
• The learned representations provide fine-grained control over recommendation aspects.
• This work paves the way for more interpretable and controllable recommendation paradigms.