Automatic Scoring System for CET-4 Compositions Based on Seq2Seq+Bi-LSTM Model
Jun Zhao, Jiulong Chen · 2020 IEEE 3rd International Conference on Automation, Electronics and Electrical Engineering (AUTEEE) · 2020
In National College English Test Band 4(CET-4), manual marking mechanism for each composition involves hard work and difficulty to guarantee the accuracy and objectivity of the scoring. It is imperative to adopt an objective, accurate and efficient automatic composition marking system. With the rapid development of artificial intelligence and deep learning in natural language processing, based on the combination of Seq2Seq and Bi-LSTM, this paper proposed a new composition scoring model for National College English Test Band 4. This model combines Bi-LSTM unit and CNN structure to realize an automatic text proof reading model. The comparison of the composition before and after proof reading will generate the resulting scores. The experimental results show that the results of this model have the highest agreement with manual scoring system compared with other classic models.