Analysis of facilitators’ behaviors in multi-party conversations for constructing a digital facilitator system

Abstract

In this paper, we analyze characteristics of facilitators from multi-party conversations. The goal of our study is to construct a digital facilitator system that supports consensus-building and management of conversation for high-quality discussions. Therefore, we need facilitator’s knowledge, behavior, and patterns to realize a good digital facilitator. As the 1st step for the purpose, we focus on a macro viewpoint of facilitators’ behaviors on conversation corpora. First, we generate a model based on a decision tree that classifies each participant into a facilitator or a non-facilitator, from conversation corpora. The classification accuracies by the decision trees were 0.642 and 0.737 for two corpora, respectively. The main purpose of the decision tree generation is to extract patterns from imaginable characteristics, namely features for the classifier. Therefore, next, we discuss behaviors of facilitators by analyzing the decision tree manually. In the analysis, we focus on two types of corpora; one is that each participant has a role, such as a project manager, and another is that each participant has no role in the conversation. We investigate the influence of the difference of the setting through the analysis. From the manual analysis, we obtained some common behaviors and some different behaviors about facilitators from two corpora.