Predicting Positive and Negative Relationships in Large Social Networks

Area of Science:

• Social Network Analysis
• Machine Learning
• Computational Social Science

Background:

• Social networks involve users expressing positive and negative attitudes, forming binary relationships crucial for analysis.
• Predicting latent (hidden) binary relationships is challenging, especially in large-scale networks.
• Existing methods struggle with accuracy as network size increases.

Purpose of the Study:

• To develop a machine learning algorithm for predicting positive and negative relationships in social networks.
• To address the challenge of decreased prediction accuracy when users have fewer common neighbors.
• To enable accurate relationship prediction in large-scale social network data.

Main Methods:

• Proposed a machine learning algorithm inspired by structural balance and social status theories.
• Implemented a segment-based training framework, dividing data by common neighbor count.
• Utilized Support Vector Machines (SVM) for prediction models within segments.
• Employed a sampling strategy to manage large datasets while preserving prediction accuracy.

Main Results:

• Prediction accuracy deteriorates when users share fewer common neighbors.
• The proposed segment-based training framework effectively handles varying common neighbor counts.
• The sampling strategy allows for efficient processing of large-scale social network data.
• The algorithm consistently outperformed traditional methods and adaptive boosting in experiments.

Conclusions:

• The developed algorithm accurately predicts latent positive and negative relationships in social networks.
• The segment-based approach and sampling strategy enhance performance on large datasets.
• This method offers a robust solution for social network analysis and relationship prediction.