TY - GEN
T1 - Comment Text Grading for Chinese Graduate Academic Dissertation Using Attention Convolutional Neural Networks
AU - Zhang, Yupei
AU - Zhou, Yaya
AU - Xiao, Min
AU - Shang, Xuequn
N1 - Publisher Copyright:
© 2021 IEEE.
PY - 2021
Y1 - 2021
N2 - Educational big data connects learning science with data science, where various educational problems are formulating into data mining tasks towards new solutions and new discoveries. This paper provides a path of automatically grading graduate academic dissertations according to the expert-given comment texts. The proposed method fed comment texts to an attention convolutional neural network consisted of an embedding layer, an attention mechanism layer, a convolutional layer, and a fully connected neural network, where the data imbalance issue was handled by data augmentations. The used comment texts were collected from 943 students spreading at 145 universities in China, where these review comments were yielded by experts to grade the academic dissertations. The results from the proposed method achieve a classification accuracy of 77% that gains 12% and 15% implementations compared to the classical convolutional neural network and the linear support vector machine. However, the result analyses show that there are many conflicts between expert-given comments and their given grades in the collected data. This study provides an automatic tool that could remove these conflicts in the dissertation review, leading to more objective dissertation grades.
AB - Educational big data connects learning science with data science, where various educational problems are formulating into data mining tasks towards new solutions and new discoveries. This paper provides a path of automatically grading graduate academic dissertations according to the expert-given comment texts. The proposed method fed comment texts to an attention convolutional neural network consisted of an embedding layer, an attention mechanism layer, a convolutional layer, and a fully connected neural network, where the data imbalance issue was handled by data augmentations. The used comment texts were collected from 943 students spreading at 145 universities in China, where these review comments were yielded by experts to grade the academic dissertations. The results from the proposed method achieve a classification accuracy of 77% that gains 12% and 15% implementations compared to the classical convolutional neural network and the linear support vector machine. However, the result analyses show that there are many conflicts between expert-given comments and their given grades in the collected data. This study provides an automatic tool that could remove these conflicts in the dissertation review, leading to more objective dissertation grades.
UR - http://www.scopus.com/inward/record.url?scp=85124961220&partnerID=8YFLogxK
U2 - 10.1109/ICSAI53574.2021.9664159
DO - 10.1109/ICSAI53574.2021.9664159
M3 - 会议稿件
AN - SCOPUS:85124961220
T3 - ICSAI 2021 - 7th International Conference on Systems and Informatics
BT - ICSAI 2021 - 7th International Conference on Systems and Informatics
A2 - Yang, Jianxi
A2 - Li, Kenli
A2 - Tu, Wanqing
A2 - Xiao, Zheng
A2 - Wang, Libo
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 7th International Conference on Systems and Informatics, ICSAI 2021
Y2 - 13 November 2021 through 15 November 2021
ER -