Cascade Prediction With Self-Exciting Point Process and Local User Influence Measurement

Abstract

Read online

With the rise and large-scale applications of social networking service, the prediction of information cascades has attracted extensive attention of researchers. User influence is an important factor affecting the dissemination of posts in online social networks. However, current studies usually take the number of users’ neighbors as their influence, and do not accurately describe the role of participating users in information dissemination. In this paper, a prediction model of information cascades in social networks is established based on the Hawkes process, and the model considers three factors, i.e., post influence, user influence and users’ response time, to describe the occurrence probability of forwarding events. In order to utilize abundant information of local network topology, we present a new method of calculating user influence, combining with semi-local centrality and local clustering coefficients. Then, a regression tree algorithm is used to determine time correction coefficients to reveal dynamic post influence, and the popularity prediction of posts in social networks is realized. Comparison experiments of different models are carried out on real-world datasets to evaluate the effectiveness and prediction performance of the proposed model, and results show that our method outperforms other counterparts.

Keywords

• cascade prediction
• self-exciting point process
• user influence
• dynamic post correction
• social network