Research On Prediction Of News Review Vote Based On Deep Learning

With the development of the Internet and the World Wide Web makes it appear that many researchers have started to study online reviews and have already made good progress.For example,according to the usefulness of online product reviews,the reviews can be sorted and then buyers can get comments with more valuable information,which makes it easier to complete purchase decisions.Moreover,most news research focused on excavating the emotional factors contained in news commentary,evaluating the quality of news commentary.Almost all news sites have set up comments and vote function nowadays,in this way high quality reviews can always get more votes.The purpose of this paper is to study the intrinsic link between news commentary and the number of votes received by news commentators.The proposed model can predict the number of votes for news review.This paper treats the number of votes for news review as the popularity of news review.In summary,this paper's contributions are as follows:(1)Using the vote count of the news review as the popularity of review,we proposed a task for predicting the popularity of news review.(2)Building a corpus about news reviews.After analyzing the domestic news websites,Toutiao and Net Ease news were selected as the sources of data.The news,news commentary and the number of votes were recorded in the corpus.(3)This paper extracts text features,sentiment features,topic distribution features and so on from news reviews according to relevant research,formulates an evaluation system for news review popularity prediction task.(4)Using machine learning methods to construct comparative tests.This paper selects three different machine learning classifiers,including Support Vector Machines,Random Forest and XGBoost to predict the popularity of reviews.The best experimental results are used as the baseline.(5)Using convolutional neural network to model this task.Pre-trained word vectors are used as the input.This paper proposes a new CNN,which builds a connect layer to concatenate the feature vector getting from pooling layer and other commentary features.And then sent to the neural network for classification,which further improves the experimental results.