Link prediction in recommender systems with confidence measures.

Area of Science:

• Network analysis
• Data mining
• Computational social science

Background:

• Link prediction is crucial for understanding network structures and dynamics.
• Existing methods often lack a robust assessment of prediction confidence.
• Information asymmetry can reduce the reliability of similarity calculations in networks.

Purpose of the Study:

• To propose a similarity confidence coefficient to improve link prediction accuracy.
• To introduce a confidence measure for quantifying the reliability of link prediction results.
• To address the underexplored aspect of confidence in network link prediction.

Main Methods:

• Development of a similarity confidence coefficient to balance similarity calculation reliability.
• Introduction of a confidence measure to quantify prediction certainty.
• Experimental validation using the Movie-Lens dataset.

Main Results:

• The similarity confidence coefficient improved prediction accuracy by correcting node similarity.
• The proposed confidence measure successfully quantified the confidence degree of link predictions.
• Confidence levels were found to be quantitatively determined by the amount of data used.

Conclusions:

• The novel confidence coefficient and measure enhance the robustness and interpretability of link prediction.
• Quantitative confidence assessment is vital for practical applications of link prediction.
• Future work can explore the application of these confidence measures in diverse network domains.