With the acceleration of globalization, the importance of business English writing in cross-cultural communication is becoming more and more prominent. This study proposes an automated assistance platform for English business writing based on grammatical analysis algorithms, which deeply integrates the RST-Style discourse parser improved by the Conditional Random Field CRF with the GloVe global semantic word vector model to solve the deficiencies of traditional methods in long-distance dependency and lexical semantic association, and introduces a sequence-to-sequence error correction model based on the replication mechanism combined with the BERT pre-training language model to optimize the semantic representation and error correction efficiency. Through multi-dimensional experimental validation, the model has an average absolute error MAE of 2.071 and a Pearson’s correlation coefficient PCC of 0.702 in the lexical articulation diagnosis task.The pairwise accuracy PRA for logical coherence diagnosis on the Accident and Earthquake datasets are 96.57% and 97.98%, respectively. The F1 value for the grammatical error detection task reaches 69.84%, which is significantly better than the baseline model. The teaching application experiments show that the mean of the total posttest score of the experimental group using the platform improves to 90.41 (58.87 on the pre-test), and the subdimensions of lexical articulation and grammatical accuracy are close to full scores of 23.16 and 24.01, respectively, and the standard deviation is significantly narrowed, which confirms the practical value of the platform in improving writing ability and teaching efficiency.
