Forecasting popularity of news article by title analyzing with BN-LSTM network

Abstract

In recent years, predicting the popularity of articles in the news has become a more urgent task for authors, online resources and advertisers. In the order of this task, we propose a new method based on the Online Deep Neural network with Bottleneck compression, what predicts the article popularity with only its headline. The proposed methodology evaluated on the Chinese and Russian language-based datasets with over than 800 000 samples in total. We describe the challenges and solutions related to the popularity prediction and the headline analysis. We show that the provided method can reach acceptable results even with different languages, news source popularity dynamics.