Information Filtering via Heterogeneous Diffusion in Online Bipartite Networks

Area of Science:

• Information Science
• Computer Science
• Network Analysis

Background:

• The internet's vast information requires effective filtering mechanisms.
• Recommender systems, particularly diffusion-based algorithms, are crucial for information navigation.
• Existing diffusion models often assume symmetry between user-object and object-user interactions.

Purpose of the Study:

• To investigate the asymmetry in diffusion processes within user-object networks.
• To enhance recommender system performance by incorporating this asymmetry.
• To improve both accuracy and diversity in recommendations.

Main Methods:

• Modifying state-of-the-art diffusion-based recommendation algorithms.
• Introducing a framework that accounts for the non-equivalent nature of user-to-object and object-to-user diffusion.
• Conducting simulations to evaluate the proposed modifications.

Main Results:

• The proposed methods outperform existing algorithms in recommendation accuracy.
• The new approaches demonstrate superior performance in recommendation diversity.
• The modifications show effectiveness in a realistic case study.

Conclusions:

• The asymmetry of diffusion in user-object networks is a critical factor for recommendation quality.
• Accounting for this asymmetry leads to significant improvements in recommender systems.
• This research offers a more nuanced and effective approach to personalized recommendations.