A Social News Propagation Prediction Method Based On No Topology Structure

With the development of network technology,the network has become a part of people's lives,and more and more people are enjoying the convenience brought by the network.People share and get all kinds of news through social networking platforms.It is very important to accurately predict the spread of information in social networks,and this issue is also widely concerned in the field of data mining.At present,most researches use social network topology and action logs during user message propagation to predict the message propagation scope.Macroscopically,the change process of message propagation at different moments is used to predict the message propagation range;microscopically,the user network topology and message propagation process are used to predict the message propagation range.In practical applications,user action logs are easy to obtain,but social network user topology and message propagation structure are not easy to obtain.Therefore,the social message prediction without topology has a wider application prospect.In this paper,three methods are proposed to predict the propagation range of messages without topology.(1)The NT-EP(non-topology message propagation)method first uses the characteristics of message propagation to decay with time to construct a weighted graph of the message propagation structure,and uses a random walk strategy to obtain multiple message propagation paths.Secondly,put the message propagation path into Bi-GRU(bidirectional gated recurrent unite),and combine the attention mechanism to calculate the propagation feature vector of the target message.Then,through the gradient descent method,the influence vector of other messages to the target message is calculated.Finally,the target message propagation vector is combined with other message influence vectors to predict the final propagation range of the message.The experimental results on the Sina Weibo and Flixster datasets show that the NT-EP method is superior to the existing social message propagation range prediction methods in terms of mean squared error(MSE),F1-score and other indicators.(2)The NTG-EP(non-topology graph message propagation)method first uses the influence between users to construct a weighted graph,and randomly extracts multiple information propagation paths from it.Secondly,according to the user's influence relationship,characterize the user's influence vector.Then,the user influence vector during the propagation of the target message is calculated according to the user's influence vector and the influence weight matrix between the constructed users.Finally,the target message propagation vector is calculated based on the user vectors participating in the target message propagation,and the message propagation range is predicted based on the target message propagation vector.The NTG-EP method requires a shorter sampling interval from the beginning of the message to the prediction of the message propagation range.The experimental results on the Sina Weibo and Flixster data sets show that the evaluation index of the mean square error of the NTG-EP method is superior to the existing social message propagation prediction methods.(3)The NT-TP(non-topology time propagation)method first constructs a weighted graph according to action logs at different moments in the message propagation process,and extracts multiple message propagation paths from the weighted graph.Secondly,according to the user's historical influence vector and the constructed weighted propagation graph,calculate the short-term influence vector of the user participating in the message propagation.Then,according to the user's short-term influence vectors at different times of the message,the message propagation vectors at different times of the message are constructed.Finally,according to the message propagation vectors at different moments of the message,they are put into the LSTM in the order of time occurrence to predict the message propagation range.Compared with NT-EP method and NTG-EP method,NT-TP has a smaller number of sampling action logs,and it takes less time to predict the propagation of messages.The experimental results on Sina Weibo and Flixster data sets show that the evaluation index of the mean square error of the NT-TP method is better than the existing social message propagation prediction methods,and the time required to predict the message propagation scope is shorter and time-sensitive Stronger.