Research on Automated Scoring Method for HSK Essays with Hybrid Features

Automated essay scoring (AES) has assisted or replaced tedious manual scoring in many occasions. However,research on the technology and application of automatic scoring for Chinese as a second language essays is still insufficient. In this paper, we extract interlanguage error features from the HSK (Chinese Proficiency Test) essay corpus, integrate these with word-level semantic features of a pre-training language model, and employ a multi-feature hybrid model to enhance the consistency and interpretability of essay scoring. Experiments show that the proposed model improves the scoring consistency by 6.77% and 5.53% compared to the baseline pre-training model and the pre-training model fusing traditional features, respectively. Meanwhile, the proposed model performs more consistently in scoring experiments on various topics, inter-language error features are negatively correlated with scores and can better assist the final decision of the model compared to traditional features, suggesting that inter-language error features are more effective for representing text information in the scoring task than traditional essay features.