Deep Reinforcement Learning Framework for Category-Based Item Recommendation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Recommender Systems

Background:

• Deep reinforcement learning (DRL) optimizes long-term user engagement in recommender systems.
• Large action spaces in DRL-based recommenders hinder sampling efficiency and accuracy.

Purpose of the Study:

• To address the challenge of large action spaces in DRL recommender systems.
• To improve recommendation accuracy and long-term user engagement.

Main Methods:

• Proposing a deep hierarchical category-based recommender system (DHCRS).
• Reconstructing the action space into a two-level category-item hierarchy.
• Utilizing two deep Q-networks (DQNs): one for category selection and one for item selection within categories.
• Introducing a bidirectional category selection (BCS) technique.

Main Results:

• Significantly reduced action space for each DQN.
• More effective capture of user preferences through item categorization.
• Outperformance of state-of-the-art methods in hit rate and normalized discounted cumulative gain.

Conclusions:

• DHCRS effectively handles large action spaces in DRL recommender systems.
• The hierarchical approach enhances recommendation accuracy and long-term user engagement.
• The BCS technique further improves the system's ability to capture user preferences.