AUI&GIV: Recommendation with Asymmetric User Influence and Global Importance Value

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Mining

Background:

• Traditional user-based collaborative filtering (CF) relies on symmetric similarity, assuming equal influence between users.
• This symmetry overlooks expert-novice dynamics and individual user importance in recommendations.
• Existing methods fail to capture the directed nature of user influence.

Purpose of the Study:

• To propose an asymmetric user influence model for more accurate recommendation systems.
• To integrate user-based directed influence with global user importance for enhanced CF.
• To improve the performance of user-based CF algorithms, especially on dense datasets.

Main Methods:

• Developed an asymmetric user influence model to quantify directed influence between users.
• Utilized the PageRank algorithm to compute the global importance of each user.
• Integrated directed and global influence values to derive final user influence scores.
• Applied these refined influence scores to enhance the traditional user-based CF algorithm.

Main Results:

• The asymmetric user influence model significantly improves recommendation accuracy.
• Global user importance values are crucial for enhancing recommendation performance.
• The proposed method demonstrates superior performance compared to existing algorithms, particularly user-based CF, on datasets with high rating density.

Conclusions:

• Asymmetric influence modeling and global importance are vital for accurate recommendations.
• The proposed integrated approach offers a substantial improvement over traditional CF.
• This research provides a more nuanced and effective method for personalized recommendations.