Hawkes process marked with topics and its application to Twitter data analysis

¹ Graduate School of Mathematical Sciences, University of Tokyo - 3-8-1, Komaba, Meguro-ku, Tokyo, 153-0041, Japan
² Japan Science and Technology, CREST - Kawaguchi, Japan
³ National Institute of Informatics - 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo, 101-8430, Japan
⁴ Tokio Marine and Nichido Risk Consulting Co. Ltd. - 1-5-1, Otemachi, Chiyoda-ku, Tokyo, 100-0004, Japan

^(a) E-mail: godamasa235@gmail.com (corresponding author)

Received: 4 August 2022
Accepted: 30 November 2022

Abstract

On Twitter, complex networks of information propagation are observed. In this article, we propose a method to model the time interval of user's tweets by combining the marked Hawkes process and the Latent Dirichlet Allocation method. We also propose a method that quantifies how individual topics in tweets excite future tweets and visualize the results as a topic-wise Hawkes graph. As an application to actual data, we analyze the tweets from the American, Chinese, and British embassies from February 1, 2020, to September 27, 2020, a period that roughly corresponds to the initial outbreak of the COVID-19 pandemic.