Research On News Popularity Prediction Methods Based On Deep Learning

Nowadays,the development of the internet is in full swing,giving birth to a variety of network applications,especially Web2.0 and the arrival of the era of big data.A large amount of network news data is used to analyze the stock market,personal finance and pay attention to national financial affairs.Network news,a simple and fast way,has gradually been favored by more and more users.Yet major news websites every day the number of financial news reports is various,quality is good and bad are intermingled,the user can not expend energy all view to obtain useful information,users tend to focus on hot news of finance and economics,the article researches on the prediction problem for hot news of finance and economics,combining with the characteristics of the news text,the paper main work is as follows:1)Financial news data fetching and pretreatment:use python network news data capture system,was designed and implemented using the crawler from Sohu news site the financial news column of fetching the following two types of news:popular and unpopular news,and by grasping for a long time,the accumulation of time to get a lot of financial news and data after the news.in order to avoid unnecessary errors,before the formation of Chinese corpus,word segmentation and stop-words filtration technologies are essential.After a series of processing,news corpus marked with hot and non-hot are finally obtained.2)Put forward the deep learning model of hot financial news:Using double two-way LSTM variation GRU helped both short-term and long-term memory neural network mechanism and focus attention to build the network news network model of prediction of the deep learning popular in Chinese Wikipedia and lab Sogou network news corpus uses Word2Vec training vectors,use the Word embedded layer(Word Embedding),said the news text words vector initialized using the words in the process of the training vectors,and constantly adjust in the training process of the model,Finally use full connection layer(Dense)hot news of finance and economics of the projections,the experimental results show that the model based on BiGRU-Attention based deep learning model predicted results is better than the other,only by simple adjustment and use GRU helped,greatly saves the program running time,text classification for deep learning provides a good foundation.