Deep Automated Text Scoring Model Based on Memory Network
- Resource Type
- Conference
- Authors
- Yang, Shiyan
- Source
- 2020 International Conference on Computer Vision, Image and Deep Learning (CVIDL) CVIDL Computer Vision, Image and Deep Learning (CVIDL), 2020 International Conference on. :480-484 Jul, 2020
- Subject
- Computing and Processing
Recurrent neural networks
Deep learning
Speech recognition
Sentiment analysis
Labeling
Knowledge discovery
Knowledge based systems
Automated Text Scoring
LSTM
Neural Network
Memory Network
Natural Language Processing
- Language
Automated text scoring system provides an effective alternative to manual scoring because of its advantages in speed and integrity of scoring. Currently in most online examination software, the scoring is only focused on the writing abilities in grammar and styles rather than the content details. To assess the content of the text answer, an auto-grading system is developed for short answered questions. It combines word weights in different aspects and uses the generated weight matrix for system to concentrate on the important words and phrases. Then it stores the knowledge base in the memory and use neural network for information fusion of the student’s answer and the weighted facts. The final score is predicted based on the fusion result. The system is tested on the dataset of online technology interview question and answers. Experiment results show that it has good generalization ability and gets accuracy rate of 83.37%, which can be used in assisting human grading while saving substantial human resources.