Prediction And Evaluation Of Information Diffusion Path Based On A Topic-oriented Relationship Strength Network

Entering the web 2.0 era,online social networks have gradually become an important channel for information dissemination.It is of practical significance to study the information dissemination in social networks.Analyzing the users' interaction behaviors and their characteristics,and then studying the information diffusion path,we can explore the information diffusion process of social networks from the micro level.Predicting information diffusion paths can predict the direction of topic diffusion,and evaluating information diffusion paths can find the key edges that play an important role in topic diffusion,so the study of information diffusion paths in social networks can help provide scientific basis for early warning and intervention of public opinion at the micro level.In this study,by analyzing the historical interaction records among users and mining the users' interaction characteristics,we realize the topic-oriented prediction and evaluation of information dissemination paths.Firstly,we use LDA topic model to realize the topic segmentation of the text content of user interaction records,analyze the topic preference of users' interactions and discover the temporal correlation of user interactions in terms of interaction frequency and interaction topics.On this basis we construct a relationship strength network reflecting user interaction preferences.Secondly,three types of features,such as user,topic and social,are mined from users' historical interaction records to achieve information diffusion path prediction cascade by cascade based on machine learning methods.The comparison experiments show that the model with the two features of topic-oriented relationship strength and path width is more competitive in the prediction of different cascades and different machine learning models.Finally,from the perspective of information dissemination,a path evaluation model based on information diffusion capacity is proposed considering the characteristics of topic preference,interaction frequency and spreading activity.The model and the baseline method(path evaluation based on network connectivity)are compared from the perspective of differentiated granularity and dissemination ability.The results show that the method based on information diffusion capacity can differentiate the paths in finer grain under different topics;SIR simulation experiments verify the effectiveness of our evaluation model in information dissemination and discover the backbone network and weak connections in the network.