Seven-Layer Model in Complex Networks Link Prediction: A Survey.

Area of Science:

• Complex networks analysis
• Network science
• Data mining

Background:

• Link prediction is fundamental to understanding complex networks.
• Existing methods lack interdisciplinary integration and a unified classification.
• Network evolution analysis relies heavily on accurate link prediction models.

Purpose of the Study:

• To survey and critically analyze existing link prediction methods.
• To propose a novel seven-layer stratified model for classifying link prediction techniques.
• To identify the most effective features and methods for link prediction.

Main Methods:

• Analysis of topological, temporal, attribute, label, weight, directional, and symbolic features.
• Introduction of a seven-layer model encompassing network, metadata, feature classification, input selection, processing, output selection, and output layers.
• Categorization of link prediction methods within the processing layer into similarity-based, probabilistic, likelihood, supervised, semi-supervised, unsupervised, and reinforcement learning approaches.

Main Results:

• Link prediction methods leveraging graph structure features demonstrate superior predictive performance.
• A comprehensive discussion of input features, evaluation metrics, comparative analyses, and datasets for various methods.
• Identification of strengths and weaknesses across different link prediction techniques.

Conclusions:

• Graph structure-based link prediction methods are highly effective.
• The proposed seven-layer model provides a structured framework for understanding and developing link prediction techniques.
• Future research should focus on interdisciplinary approaches and advanced feature engineering for complex networks.