Extracting the information backbone in online system.

Area of Science:

• Computer Science
• Information Science
• Network Analysis

Background:

• Information overload is a significant societal challenge.
• Recommender systems aim to filter irrelevant information.
• Existing research primarily focuses on algorithm performance (accuracy, diversity) rather than network topology.

Purpose of the Study:

• To investigate the impact of user-object bipartite network topology on recommender systems.
• To address the issue of redundant and misleading information within these networks.
• To develop novel algorithms for improving recommender system effectiveness and efficiency.

Main Methods:

• Analysis of information redundancy and misleading features in bipartite networks.
• Design of link elimination algorithms to refine network topology.
• Development of a hybrid approach combining time-aware and topology-aware link removal.
• Extraction of essential network backbones for recommender systems.

Main Results:

• Identified that certain network links provide redundant or misleading information.
• Demonstrated that eliminating specific links enhances recommendation performance.
• Proposed a hybrid method to extract a network backbone.
• Achieved improved recommendation accuracy and diversity.

Conclusions:

• Network topology significantly influences recommender system performance.
• A 'less is more' approach by removing redundant links is beneficial.
• The proposed hybrid method improves both the effectiveness and efficiency of recommender systems.
• Reduced computational time and enhanced recommendation quality are key outcomes.