The Research Of Popularity Prediction Algorithm Of The Message On Social Media

In recent years,with the popularization and development of the mobile Internet,the social media which provides individuals with free creation and content sharing has entered on a new stage,and it has become an indispensable part of people's daily life.There is mass information attracting users' attention through being browsed,forwarded,and given thumbs-up on social media.Meanwhile,information mentioned above generates huge network flow,and its popularity distribution is not uniform,which means that only a small amount of content attracts much attention.Therefore,predicting the popularity of the content on social media has become a valuable research issue,which is of great significance for online advertising,online marketing and so on.This thesis introduces the research status of the prediction algorithm of the content popularity on social media at home and abroad.By analyzing the features related to forwarding and applying the Susceptible Infected model,we propose a solution to solve the problem of predicting the content popularity on social media.The main work of this thesis is listed as follows:1.A logistic regression classifier based on input parameters which are the fans number of the publisher who post the tweet,the release time,the first forwarding time,the early forwarding time interval,the maximum fans number of early forwarding users,and the average number of early forwarding users to classify tweets and determine whether it belongs to the class in which its forwarding volume is Top20 high.And the final classification accuracy is 0.77.Analyzing the set of the above 6 features and the forwarding volume through PCA,it is found that after compressing the feature data into2 dimensions,and calculating the absolute value of the cosine similarity of the projection vector of each feature to the projection vector of the forwarding volume,3features stands out for their values are approximated to 1.Meanwhile,the single feature prediction accuracy of the 3 features is higher than the others.2.A popularity prediction algorithm based on the disease propagation model is proposed to perform numerical prediction of the future forwarding volume of the tweet.The disease propagation model is modified,where the influence of time on users' interest to forward is added,represented by the time decay function.When applied to the prediction of the forwarding volume of tweets,the time decay function is specialized as a power function.The objective function is the square residual of the predicted valueof the algorithm and the actual amount of forwarding,and the parameters are trained by the LM algorithm.The algorithm of this thesis is compared with the benchmark algorithm on the twitter dataset.The experimental results show that the popularity prediction made by algorithm of this thesis is more accurate than the benchmark algorithm(the MRE of the prediction value of Top25 tweets is improved by 12.4%),what's more,it is able to simulate the change of popularity.