Hot Articles Forecasting System Based On The Headline Analysis With Deep Learning

Forecasting popularity of the articles become one hot task for the news media industry and the research community have provided several prediction systems,based on popular trends,metadata and users reaction.However,when the headline has a fundamental relationship with the popularity of the article,there is no proper system,which can use only the title to predict the possible views amount.This theses work proposes a new Hot Articles Forecasting System based on the headline analysis with Online Deep Neural Network architecture and Bottleneck compression technique,which can make an efficient real-time prediction of the article popularity and automatically learn new data.Besides,this work introduces three new datasets based on the Chinese and Russian languages,with over than 800 000 samples in total.In advanced,this thesis provides the corporation of the formal and informal language style headlines in the case of the popularity prediction for the Chinese language.This work also investigates the challenges and provides solutions related to the popularity prediction and the headline analysis.This work will show that the HAFS can provide prediction results with different languages in the case of changing activity and trends.