Microblog Forwarding Prediction Based On Deep Learning

As the largest social media in China,Sina Weibo has a huge amount of data and users.The forwarding prediction of Weibo has become a research hotspot in the current academic circles.At present,most of the microblog forwarding research is based on the traditional model.The model only uses specific data attributes or uses statistical features similar to TF-IDF for training.It does not extract the features of the microblog semantic level.For the above situation,this paper uses the depth model.Semantic feature extraction,and proposed a self-attention Attention mechanism,the main work is as follows:Visualize the data set for the Sina Weibo dataset.Through the visual analysis of the data in the three dimensions of forwarding amount,praise and comment volume,the user is screened,and the appropriate data set is provided for subsequent modeling and the key factors affecting the microblog forwarding are found in the data set.The detailed feature engineering of the data set is carried out to extract the features needed in the traditional model modeling process,and the features are selected by random forest.Finally,XGBOOST is used to train and complete the modeling of the traditional model.In the semantics of Weibo,the RNN network which is good at processing time series data in deep learning is used to extract the semantic features of Weibo.The results obtained by different kinds of RNN networks are compared during the training,and strategies to prevent overfitting are added.For the RNN unit,a self-attention Attention mechanism is proposed to make the depth model have different training weights for different words and improve the feature extraction ability of the RNN network.Finally,the depth model is fitted with the traditional model,so that the network can not only carry out traditional user modeling but also understand the microblog content,and improve the accuracy of microblog forwarding prediction.After the experimental processing,the final model prediction results are compared with the traditional model prediction results,which proves that the deep network with self-attention mechanism can really help the traditional network to improve the accuracy and improve the final effect of the model.