A Study on Automatic Scoring and Diagnostic Feedback Generation of Vocational College English Application Writing Based on a Large Language Model

Abstract

This paper examines the use of large language models to construct an automatic scoring and diagnostic feedback generation system for vocational college English application writing. First, a corpus of vocational college English application writing texts covering multiple genres is constructed. Based on manual scoring and multidimensional feature annotation, a scoring rubric aligned with vocational college English teaching practice is developed. Second, a finely tuned pre-trained language model is used to implement an automatic scoring module and a diagnostic feedback generation module. The scoring module adopts a BERT-based architecture for discriminative scoring, while the feedback module uses a generative model to produce structured diagnostic comments. Experimental results show that the mean weighted consistency coefficient between automatic and manual scoring reaches 0.87. Diagnostic feedback is positively evaluated in terms of content accuracy, personalization, and teaching usefulness. The system can generate overall evaluations, dimensional diagnoses, and targeted improvement suggestions for applied writing. The study provides an effective intelligent auxiliary tool for English writing instruction in vocational colleges and supports scalable formative assessment.