A Subsequent Speaker Selection Method for Online Discussions Based on the Multi-armed Bandit Algorithm

Abstract

This paper proposes a method to select subsequent speakers in an online discussion, which is one of the important functions of facilitators, using the multi-armed bandit algorithm. Bandit algorithms can be applied to speaker determination by considering each participant as an arm of a slot machine and a facilitator as a player. We define a “discussion score” to evaluate each post, and it is then considered to be equivalent to the reward of the slot machine method. The discussion score of each post is defined based on the following three metrics: (1) Whether the post helps to settle a discussion or not. (2) How interested are the other participants in the post (3) The intention of the post. To consider conflict between participants, our method classifies the participants into groups and determines the next speaker based on clustering results. We demonstrate that our method can select participants who posted good ideas and opinions and promote participants to engage other participants by using questionnaires.