Automatic Short Answer Grading with Rubric-based Semantic Embedding Optimization

• Item type: SIG Technical Reports(1)
• 公開日: 2022-02-14
• タイトル: Automatic Short Answer Grading with Rubric-based Semantic Embedding Optimization
• タイトル / 言語: en
• タイトル: Automatic Short Answer Grading with Rubric-based Semantic Embedding Optimization
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Kyushu Universtiy
• 著者所属: The National Center for University Entrance Examinations
• 著者所属: Kyushu Universtiy
• 著者所属(英): en
• 著者所属(英): Kyushu Universtiy
• 著者所属(英): en
• 著者所属(英): The National Center for University Entrance Examinations
• 著者所属(英): en
• 著者所属(英): Kyushu Universtiy
• 著者名: Bo, Wang; Tsunenori, Ishioka; Tsunenori, Mine
• 著者名(英): Bo, Wang; Tsunenori, Ishioka; Tsunenori, Mine
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Large-scaled encoders such as BERT have been actively used for sentence embedding in automatic scoring. However, the embedding may not be optimal due to non-uniform vector distribution. By conducting fast contrastive learning, methods like SBERT got better semantic embeddings and were actively used in textual similarity datasets. However, the cost to obtain the similarities limits its application to automatic grading. In this paper, we propose a method of calculating similarity from the rubric to perform contrastive learning for a better semantic embedding. We conducted extensive experiments on 60,000 answer/question data for three independent questions. The experimental results show that the proposed method outperforms all baselines in terms of accuracy and computation time.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Large-scaled encoders such as BERT have been actively used for sentence embedding in automatic scoring. However, the embedding may not be optimal due to non-uniform vector distribution. By conducting fast contrastive learning, methods like SBERT got better semantic embeddings and were actively used in textual similarity datasets. However, the cost to obtain the similarities limits its application to automatic grading. In this paper, we propose a method of calculating similarity from the rubric to perform contrastive learning for a better semantic embedding. We conducted extensive experiments on 60,000 answer/question data for three independent questions. The experimental results show that the proposed method outperforms all baselines in terms of accuracy and computation time.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11135936
• 書誌情報: 研究報告知能システム（ICS）; 巻 2022-ICS-205, 号 11, p. 1-7, 発行日 2022-02-14
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-885X
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会