Link Prediction in Weighted Networks: A Weighted Mutual Information Model

Area of Science:

• Data Mining
• Network Analysis
• Knowledge Discovery

Background:

• Link prediction is a key challenge in data mining and knowledge discovery.
• Existing methods primarily focus on unweighted networks, neglecting the impact of link weights.

Purpose of the Study:

• To develop a novel weighted model for link prediction in undirected and weighted networks.
• To enhance the accuracy of link prediction by incorporating link weights into local network structure analysis.

Main Methods:

• A weighted model based on mutual information of local network structures was developed.
• Link weights were utilized to improve the distinguishability of candidate links.
• Empirical experiments were conducted on four real-world weighted networks.

Main Results:

• The proposed weighted model significantly outperformed traditional unweighted indices.
• The method also demonstrated superior performance compared to existing typical weighted indices.
• Analysis provided insights into the role of weak ties and the potential for predicting link weights.

Conclusions:

• The developed weighted model offers a more accurate approach to link prediction in weighted networks.
• This research contributes to the advancement of algorithms for link prediction in complex weighted network structures.