Automatic evaluation and enhancement of reading strategies in English reading comprehension based on the BERT model

Abstract

The BERT (Bidirectional Encoder Representation from Transformers) model has developed rapidly in recent years. Applying it to English reading comprehension can help with automated evaluation of reading strategies. This article aims to use the BERT model to help students improve their reading comprehension abilities. Firstly, reading comprehension articles from the middle and final English exam papers of A High School in W City from 2010 to 2020 were collected to construct a dataset, and the collected data was preprocessed, mainly by standardizing and enhancing the paragraphs of the articles. Next, the BERT model was constructed to precisely extract paragraphs from reading comprehension articles through BERT’s two-dimensional encoding ability, thereby improving the accuracy of reading comprehension. Then, based on the BERT model, research was conducted on reading strategies. By comparing the vocabulary and sentence structures used in reading comprehension paragraphs and comparing them with the predicted results of the BERT model, the reading strategies used were determined and evaluated. The experiment showed that the evaluation indicators of the English reading comprehension model constructed based on BERT were higher than those constructed by other methods, with precision, recall, and F1 values of 96.5%, 93.8%, and 94.7%, respectively; the average accuracy of the model studied in this article for evaluating and classifying these eight reading strategies was 97.17%, which was 12.65% higher than the model constructed based on Transformer. The BERT model, as a training tool for improving reading strategies, can help students modify their reading strategies based on the feedback and guidance of the BERT model after undergoing extensive reading comprehension training, thereby improving their reading speed and comprehension level.