Optimizing online social networks for information propagation.

Area of Science:

• Information Science
• Computer Science
• Social Network Analysis

Background:

• Information overload is a significant challenge for online users.
• Recommender systems, particularly adaptive social recommendation, aim to mitigate this by optimizing social network connections for information propagation.
• Existing models often assume uniform user activity, which is unrealistic.

Purpose of the Study:

• To address the limitations of uniform user activity assumptions in adaptive social recommendation.
• To propose and validate a more realistic multi-agent model that incorporates heterogeneous user activity frequencies.
• To improve the efficiency and accuracy of information propagation in social networks.

Main Methods:

• Empirical analysis of online user activity distributions revealing a heterogeneous, power-law distribution.
• Development of a multi-agent model reflecting this power-law distribution of user activity.
• Design of a novel similarity measure that incorporates user activity frequencies.

Main Results:

• Previous social recommendation methods suffer from significant information propagation delays due to inactive connections.
• The proposed similarity measure effectively addresses delays caused by inactive leaders.
• The new approach significantly shortens average information propagation delay and enhances recommendation accuracy.

Conclusions:

• User activity in online social networks is heterogeneous, not uniformly distributed.
• Adaptive social recommendation models must account for this heterogeneity for optimal performance.
• The proposed activity-aware similarity measure offers a substantial improvement over existing methods for social recommendation.