PREDICTING ONLINE NEWS ARTICLE POPULARITY ACROSS SOCIAL PLATFORMS USING MACHINE LEARNING A MEDIA ANALYTICS APPROACH

Abstract

In a fast-paced publishing environment of digital media, reliability in forecasting popularity in online news articles is critical in making a move to ensure the achievement of a larger audience in media consumption and contact. In this paper a media analytics system is suggested based on machine learning where it can pre-label news stories on the internet to be popular or not popular using just metadata and linguistic features. The work is based on the use of the Online News Popularity dataset of the UCI Machine Learning Repository, which consists of more than 39 thousand articles and about 60 predictive attributes, such as text information, metadata on publishing, and sentiments using keywords. Synthetic Minority Over-sampling Technique (SMOTE) has been used to cope with class imbalance among non-popular and popular articles. Five algorithms of supervised learning were analyzed, including Logistic Regression, Random Forest, Support Vector Machine (RBF), Multilayer Perceptron and XG Boost, with and without balancing. The XG Boost classifier tuned by Grid Search CV showed the best result with F1-score: of 0.697 and 69.8 percent accuracy. The analysis with the feature importance indicated that articles that concerned entertainment and technology and were released on weekends and supplemented with multimedia were prone to go viral. These results show the potential importance of the inclusion of AI-based predictive analytics into the workflow of journalism. Through those models, newsrooms will be able to implement the principles of data-driven editorial choices, as well as plan the news releases based on the publication schedule and enhance the activity rates of the readers in the continuously growing competitive mass communication space.