An Approach for Predicting the Popularity of Online Security News Articles

Abstract

Online social platform has become the main way for people to access information and the main platform for news delivery. Journalists and editors face with the task of determining which article will receive a significant amount of user attention. The reasons that impact on the popularity of news articles are varying from complex factors including the category of news articles, author and many more. To cast the popularity prediction, we extract features from title, summary and publication date. Then a novel framework composed of four parts is designed to predict the popularity. We apply a recurrent neural network model called Gated Recurrent Unit(GRU) to our framework, experiments on our dataset including about 25939 Chinese security news articles in China’s websites and 7654 English security news articles and blogs in foreign countries. The result of experiment demonstrates that our framework outperforms the baseline model. The average mean absolute error(MAE) of Chinese security news articles is reduced by 54.6315% and the average root mean square error(RMSE) of English security news is decreased to 0.21677.