Text Error Correction Method in the Construction Industry Based on Transfer Learning

Text error correction is of great value in the review of texts in the construction industry. For construction industry texts, which are compound texts with multi-domain proper nouns, the lack of labeled data leads to poor error correction algorithms based on deep learning. For this reason, this paper proposes a text error correction method in the construction industry based on transfer learning. Based on the pre-trained BERT model, we transfer some parameters to the target error correction model after unsupervised training by unlabeled related field dataset, and then retrain the model through the training samples of the construction document corpus dataset to obtain better error correction effects. Meanwhile, we dynamically adjust the pre-training task in transfer learning to improve the performance of the word order correction task. Experimental results show that our proposed model has higher precision rate, recall rate and lower false positive rate in the error correction task than other models.