Front-page News Prediction And Application Research Of Authoritative Newspapers

As the mouthpiece of a political group,authoritative newspapers are often regarded by international media and international political research organizations as the "wind vane" for observing the current political development of political groups.With the rise of machine learning and natural language processing technologies,the research methods of authoritative newspapers have gradually shifted from high-cost manual review and judgment to efficient and fast machine representation and prediction.In order to identify important news with special significance from the authoritative newspapers efficiently and accurately,and further find the possible policy adjustment signals of current political groups based on this result,this thesis makes some contributions in the following three aspects:Firstly,this thesis proposes a front-page news prediction method(Top NText)based on the node centrality of the complex network,which solves the news recommendation problem by ranking news importance.The essence of the recommendation process is the sorting process of the importance of news.In order to realize the sorting of the importance of news,this thesis introduces a complex network and a node centrality sorting algorithm.In the node ranking link,this thesis also combines the H index and the Page Rank ranking algorithm to give a new centrality ranking algorithm H-Page Rank.The experimental results show that the news network constructed in this thesis has the general properties of complex networks,and the Top NText model using H-Page Rank centrality is significantly better than other methods in the Top-N ranking of news.Secondly,this thesis proposes a front-page news prediction method(Stack Text)based on the multimodal information overlay network,which improves the classification effect by expanding the source of news features.The classification process is the process of predicting the current sample category based on the potential relationship between the characteristics of the training sample and the category.The Stack Text model uses weighted random sampling to balance samples to alleviate sample imbalance.At the same time,the Stack Text model introduces more types of feature information of news and designs a multimodal classification model to deeply mine the relationship between features and categories.The experimental results show that the weighted random sampling module in the Stack Text model can effectively alleviate the negative impact of unbalanced news samples,and the Stack Text model can effectively solve the problem of two-category prediction of front-page news of authoritative newspapers.Finally,this thesis proposes a policy adjustment signal discovery method(PAIOL)based on front-page news predictions of authoritative newspapers,which reveals the underlying relationship between front-page news and policy adjustment.The "People's Daily" news dataset can be abstracted as a data source of bounded streaming data.Using the classification results of daily streaming data to detect concept drift and explaining the concept drift through concept adjustment of the data source,is a good way to discover possible policy adjustment signals.The experimental results show that the PAIOL model can effectively detect policy adjustment periods by day and year with sharp policy adjustments,and can effectively detect policy adjustment signals.