Method to Predict Bursty Hot Events on Twitter Based on User Relationship Network

Abstract

In the field of hot event prediction on online social networks, not considering user information leads to poor prediction effect. In this paper, a novel method that considers the behaviors and characteristics of users is proposed to identify and predict suspected bursty hot events. First, the keywords in each tweet are extracted and divided into different sets according to part of speech, and then similar topics are clustered according to semantic similarity. Second, the growth rates of topics are monitored in the sliding timestamp and the suspected bursty hot events are marked. Then, a user relationship network is constructed based on the information of the registered users on Twitter. Finally, according to the propagation trend of suspected bursty hot events in the network, the quasi-burst hot events are marked and sorted in descending order. Experimental results show that only using the historical re-tweeting behavior of users as the judgment basis to predict the current re-tweeting probability of users will lead to the phenomenon of error cascading, while taking the information of users into account can effectively improve the prediction performance. Compared with the existing methods, the proposed method improves the prediction precision rate by 27.38%, accuracy rate by 23.49%, and recall rate by 20.16%, demonstrating that it can predict bursty hot events effectively.

Keywords

• Hot event prediction
• suspected bursty hot events
• semantic similarity
• user relationship network