The Prediction Implementation Of Weibo Retweets Based On User Influence And Content Interest Features

In recent years,the popularity of the Internet has brought a vigorous development of social networking applications.In the social networking applications,users can use a variety of ways to get the latest information,exchange the information,and participate in the discussion about the interesting events anytime.This new information exchange mode greatly shortens the time from event production to diffusion.Therefore,it is significant to predict the rapid dissemination of events on the social platform which has massive information content and large user groups.In this thesis,the forwarding of weibo from interest characteristics and user influence has been systematically studied.The work of the dissertation is partly supported by the National Natural Science Foundation of China(61172072)and Academic Discipline and Postgraduate Education Project of Beijing Municipal Commission of Education.The main contents are as follows:The extraction method of the weibo content interest is analysed.According to the characteristics of weibo short text which are collected by Sina API,two schemes on extracting the characteristics of short text interest are proposed.In the first scheme,the LDA model is constructed,and the distribution probability of each weibo interest characteristic is obtained by Gibbs sampling.In the second scheme,the modified short text TF-IDF method is used to introduce the length of the text into the consideration of the word weight.By using the weight with different characteristics,the interest characteristics of each weibo are extracted.Finally,the LDA model on short text interest feature extraction is selected by comparing the results of perplexity experiment.A calculation model on influence value of weibo users based on the interest is built.In this thesis,a new algorithm which can calculate the users' influence value based on user interest is proposed according to the diversity of weibo social network.Firstly,the interest characteristics of users are extracted by the LDA model.After constructing the users'relationship network under the specific subject,the users'influence ranking under the specific interest is obtained.In the process of calculating the users influence value,a new concept of weibo user spread-rate and interest similarity which improves the accuracy of the ranking calculation is proposed.Finally,compared with the classic PageRank algorithm,advantages of interest users influence model are demonstrated by Spearman's correlation coefficient,which solves the unicity problem of calculating the users influence based on network structure.A weibo forwarding prediction method based on the interest and the influence is achieved.In this thesis,the principle of predicting the weibo event forwarding is analyzed.The prediction model by using BP neural network is proposed and built.Then,the simulation experiment is carried out.In the experiment,50K pieces of data that contain the weibo interest characteristi,user influence value and user attribute is used as the input vectors of BP neural network prediction model.The whole model system is trained and tested by vectors.Finally,the accuracy of the prediction results can be maintained at around 85%.The forecast results of prediction model is analyzed by two different models with ROC,which leads to get the conclusion that the weibo event forwarding can be predicted by the interest and user influence model.