Prediction For Popularity Of Events In Social Network Based On Subject Relationship Emotion

As the Internet and network public opinion on it becomes much more important in our daily life,prediction for network public opinion is also becoming important.This paper measures depth and width of event propagation in social network by predicting social networking event popularity.This is of great significance no matter from a business perspective or from a research perspective.According to the characteristics that social networking event popularity is difficult to predict,from the perspective of social networking text content,by analyzing relationship emotion feature among subjects,this paper constructs a time series model of social networking event prediction,and finally realizes social networking event popularity prediction,such as number of reposting,thumbs up or comment.In this paper,experimental data is crawled from texts in social network.This paper crawls data from Facebook and Twitter networking platform,by designating time interval,3 hours,6 hours,12 hours,24 hours,1 day,3 days,7 days after text published is set as specific inspection intervals for prediction,and this paper carries out the prediction and acquisition after these intervals.This paper uses Wikidata dataset and CRF model to identify the entities in the text,where the PER and ORG entities in them are the objects that need further study.This paper uses Wikidata dataset and the LSTM model to extract entity relationships between subjects in text.By mapping the relationship types to emotions between subjects,the relationship emotion matrix in the text can be obtained.On this basis,through the analysis of the relationship emotion matrix,the relationship emotion feature vector of the text can be obtained.Features include number of characters,the proportion of characters with emotion relation,relationship emotion type,relationship emotion quantity,the degree of relationship emotion cross,relationship emotion predictability based on relationship emotion,relationship emotion intensity,the degree of relationship emotion diversification,subject influence,number of history thumbs up,number of history comment,number of history repost.According to the feature extraction steps mentioned above,feature values are extracted from the original corpora material.By using Gradient Boosting Decision Tree as the prediction model,the features are inputted to the prediction model,and set the popularity of the corresponding moment as output,the prediction model is trained and optimized,so that the aim of popularity prediction is achieved.In the process of experiment,this paper trains the data through Gradient Boosting Decision Tree model,and a regression model to predict the event popularity of social network is got.Through this model,the popularity of social networking event is predicted.The prediction results show that the accuracy of the model is 95%,which means that the prediction task is well accomplished by this paper.