C-RBFNN: A user retweet behavior prediction method for hotspot topics based on improved RBF neural network

Abstract

Abstract The prediction of user behavior plays an important role in perceiving the popularity of a topic and the changes of public opinion on the Internet. In this paper, focusing on user retweeting behavior, we propose a user retweeting prediction method for hotspot topics using fuzzy theory and neural network algorithm. Firstly, RBF (Radical Basis Function) neural network is used to model users retweeting behavior, considering that neural network can be effective in simulating the non-linear relationships among complex behaviors of users. Compared with traditional neural networks, an RBF neural network has the advantages of fast convergence and local approximation to eigenvalues when dealing with large - scale network topic data and does not suffer from the problems posed by local minimum. Besides, time-decay function is introduced to make RBF neural network can adapt to dynamic changes of various influence factors in the social networks. Secondly, we introduce cloud theory in fuzzy mathematics to optimize the activation function of hidden layers of RBF neural network because of the uncertainty of the mapping between user attributes and retweeting behavior, and then we propose the C-RBF neural network (Cloud-RBFNN), which makes the method cannot only fully express the fuzziness and randomness of user retweeting behavior, but also has good approximation ability for nonlinear relationships. Finally, because the characteristics of user retweeting behavior can change over time, changes of topic development trend are obtained by using a discrete-time method and analyzing the number of users who participate in a topic in different time periods. Experiments show that the method can accurately predict the user retweeting behavior and dynamically perceive changes in hotspot topics.