Predicting Key Events in the Popularity Evolution of Online Information

Area of Science:

• Information Science
• Data Science
• Computational Social Science

Background:

• Online information popularity exhibits dynamic evolution with distinct key events: burst, peak, and fade.
• Predicting these events is crucial for applications like recommendation systems, online marketing, and rumor containment.
• Existing methods face challenges in identifying these events across diverse popularity patterns and in achieving timely predictions.

Purpose of the Study:

• To propose a novel prediction task focused on forecasting the timing of burst, peak, and fade events in online information popularity.
• To address the challenges of high variation in popularity evolution and the short timeframes in which these key events occur.
• To develop a robust and prompt prediction solution for key popularity events.

Main Methods:

• A simple moving average is employed to smooth variations in popularity data.
• A universal method is developed to identify burst, peak, and fade events across various popularity evolution patterns.
• Feature engineering and correlation analysis are used for feature selection to identify impactful predictors.
• A machine learning model is trained on selected features for prediction.
• A novel evaluation metric is designed to assess both accuracy and promptness of predictions.

Main Results:

• The proposed method effectively identifies key events in popularity evolution across different patterns.
• Feature selection enhances prediction accuracy by removing irrelevant and redundant features.
• The developed prediction solution demonstrates superiority in experimental and comparative analyses.
• The new evaluation metric effectively balances prediction accuracy and promptness.

Conclusions:

• The study presents a superior prediction solution for key events in online information popularity.
• The developed methods successfully address the challenges of popularity variation and timely event prediction.
• Accurate and prompt prediction of popularity events has significant implications for various online applications.