| 講演抄録/キーワード |
| 講演名 |
2025-03-19 15:40
Siamese Network-based Answer Similarity for Automatically Scoring Handwritten Very Short Answers ○Tuan Nam Ly・Hung Tuan Nguyen・Masaki Nakagawa(TUAT) PRMU2024-63 |
| 抄録 |
(和) |
(まだ登録されていません) |
| (英) |
This paper proposes a Siamese network-based answer similarity evaluation method to improve the automatic scoring of a single or a few character answers. The proposed Siamese network consists of twin fully convolutional networks followed by two fully connected layers. It takes two answers as input and outputs the similarity between them, and it can be trained end-to-end by a loss function. The Siamese network is to be used in the automatic scoring mechanism that computes the similarity between a student's handwritten answer and the corresponding expected values and then scores the answer as correct, incorrect, or rejected. We also propose a method for sampling the training data to train the Siamese network for automatic scoring. We conducted experiments on a collection of handwritten answers from elementary school students, consisting of 37,500 Japanese very short answers. The extensive experiments demonstrate the superiority of the proposed method over the previous techniques of perfect matching and the content-based similarity-based methods. The ablation studies also verify the generalization of the proposed method and its effectiveness for few-shot automatic scoring. |
| キーワード |
(和) |
/ / / / / / / |
| (英) |
Automatic Scoring / Siamese Network / Deep Learning / / / / / |
| 文献情報 |
信学技報, vol. 124, no. 445, PRMU2024-63, pp. 100-105, 2025年3月. |
| 資料番号 |
PRMU2024-63 |
| 発行日 |
2025-03-11 (PRMU) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
PRMU2024-63 |