A personalized reinforcement learning recommendation algorithm using bi-clustering techniques.

Area of Science:

• Computer Science
• Artificial Intelligence

Background:

• Recommender systems (RSs) are crucial for navigating online data but often provide static recommendations.
• Traditional RSs struggle to adapt to evolving user preferences, leading to suboptimal user experiences.
• Reinforcement learning (RL) offers dynamic personalization but faces challenges with computational expense and local pattern extraction.

Purpose of the Study:

• To develop a novel recommendation algorithm integrating biclustering and reinforcement learning (RL).
• To enhance the efficiency and personalization of recommender systems by addressing computational costs and improving pattern recognition.
• To investigate the optimal biclustering algorithm for RL-based recommendation tasks.

Main Methods:

• Integration of biclustering techniques with a reinforcement learning (RL) recommendation framework.
• Utilizing biclustering to create an efficient environment for RL agents, reducing computation.
• Employing biclustering for local pattern extraction to improve RL agent learning.
• Experimenting with eight state-of-the-art biclustering algorithms.
• Introducing a novel strategy for predicting item ratings within the RL framework.

Main Results:

• The proposed algorithm demonstrates reduced computational cost and enables dynamic recommendations.
• Biclustering enhances RL agent learning by identifying locally associated patterns.
• Evaluation on three movie datasets (ML100K, ML-latest-small, FilmTrust) shows promising results.
• The approach achieves improved personalization, diversity, novelty, and reduced intra-list similarity compared to existing methods.

Conclusions:

• The integration of biclustering and RL offers a significant advancement in recommender system technology.
• The proposed method effectively addresses the limitations of traditional and pure RL-based recommenders.
• This approach provides a computationally efficient and highly adaptive solution for personalized recommendations.