Research on a Text Quality Assessment System based on BERT Deep Learning

Traditional composition correction is often manually graded by teachers according to certain standards, which will be subjective. Western languages, represented by English, have widely used mature automatic composition evaluation systems in IELTS, TOEFL and other language tests. At present, the Chinese composition automatic evaluation technology is in a rapid development stage, and there are few Chinese composition automatic evaluation systems that can be applied on a large scale. Therefore, with the development of NLP, it is necessary for Chinese educators and learners to use Deep Learning technology to improve the intelligence of Chinese language assessment. Due to the lack of the assessment grade corpus of primary school students’ composition, we have established a corpus of nearly 210,000 compositions manually graded by experts. Through comparative analysis of five kinds of models closely related to neural networks, it is finally found that BERT achieved the best result, up to 45.52%. Therefore, we believe that the BERT Deep Learning model, which is easy to deploy and model, can be used to assist language educators, language test organizers and language learners in the intelligent and auxiliary grade evaluation of primary school students’ compositions.