Online discussion threads as conversation pools: predicting the growth of discussion threads on reddit

Sameera Horawalavithana, Nazim Choudhury, John Skvoretz, Adriana Iamnitchi

Abstract: This paper proposes a data-driven method that forecasts groups of topic-related, overlapping, online conversation trees. Our method is generative: given a group of original posts, it generates the resulting conversation threads with timing and authorship information. We demonstrate using two large datasets from Reddit that the microscopic properties of such groups of conversations can be accurately predicted when starting from the original posts, without knowledge of the intermediate reactions to such posts. We show that our solution significantly outperforms competitive baselines in terms of predicting the conversation structure and user engagement over time. Potential benefits of this solution include the evaluation of intervention strategies to limit disinformation.