Experimental interpretation of adequate weight-metric combination for dynamic user-based collaborative filtering.

Area of Science:

• Data Science
• Machine Learning
• Recommender Systems

Background:

• Recommender systems are crucial for personalized experiences but face challenges with subjective preferences and performance variations.
• Existing methods often use static parameters, limiting adaptability in dynamic environments.

Purpose of the Study:

• To propose diversified similarity measurements for enhancing recommendation performance.
• To investigate the impact of dynamic parameter generation and significance weighting on user-based collaborative filtering.

Main Methods:

• Examined user-based collaborative filtering, measuring item preference probabilities.
• Verified the test item bias phenomenon and analyzed neighbor counts.
• Implemented dynamic, user-wise parameter generation excluding the item of interest.
• Incorporated significance weighting to analyze user-neighbor similarities.

Main Results:

• Validated the test item bias phenomenon in similarity equations.
• Demonstrated that dynamic parameter generation improves reliability and real-time compatibility.
• Identified effective combinations of similarity weights and performance metrics through fine-tuned neighborhood inspection.

Conclusions:

• The proposed approach uniquely combines significance weighting and test-item bias mitigation.
• Diversified similarity measurements with dynamic parameters offer a more robust and accurate recommender system architecture.