Automatic evaluation of group discussion based on multi-modal interpretation

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K12110
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 62030:Learning support system-related
• Research Institution: Kyushu Institute of Technology
• Principal Investigator: Shimada Kazutaka 九州工業大学, 大学院情報工学研究院, 教授 (50346863)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: マルチモーダル / 自然言語処理 / 議論分析

Outline of Final Research Achievements

In education, active learning, such as PBL, in which multiple people discuss issues for which there are no clear answers, has gained importance in recent years. We have studied the methods for understanding multi-party discussions and the summarisation techniques. On the other hand, evaluating debate discussions is necessary in education. However, it is not easy for evaluators to assess the quality and content of debates. In this research project, we created a dataset for evaluating the quality of debate and proposed a multimodal estimation model for estimating it. We also proposed and evaluated methods for various elemental techniques related to discussion analysis, not only on the created data but also on existing datasets (AMI corpus and our corpus created in the past).
In this research project, we created three datasets, and the datasets are available on the web.

Free Research Field

自然言語処理

Academic Significance and Societal Importance of the Research Achievements

本申請課題で作成したデータは基本的にすべて無償で申請者のWebページに公開している．これらのデータは関連研究者が自由に利用することができ，学術的な意義がある．
本申請課題で対象としている議論の評価は人間でさえも評価がぶれ，公平性などの様々な問題が生じる．この問題に対して，機械による客観的な評価が可能であれば，一定の意義がある．これは，デイベートや小論文などの自動評価という観点で社会的な意義がある．