Research On Readability Of English Text Based On Deep Learning

Reading ability is an important part of language acquisition.Because people’s reading level and comprehension ability are different,It is a time-consuming and laborious task for different language learners and teachers to find text materials suitable for their own text difficulty level requirements.so there is a large amount of demand for measuring the readability of the text.This paper summarizes the research status of text readability,analyzes the measurement results of text readability,and introduces the measurement method of English text readability based on deep learning for the limitations of existing text readability measurement methods.Designed to train a readable metric model that analyzes cross-domain text,versatility,automation,and performance through learning representation.The main research work of this paper is summarized as follows:(1)An English text readability metric based on hybrid network model is proposed.Traditional formulas and machine learning-based metrics for text readability metrics rely too much on the experience of artificial experts to extract features,limit the practicability of their usefulness,and the more versatile features of the text readability metrics used with the extraction.The larger the number,the more difficult it is to manually extract deep features,and it is easy to introduce irrelevant features or redundant features,resulting in a problem of model performance degradation.This paper introduces the concept of hybrid network model in deep learning.By combining convolutional neural network and two-way long-term memory network and attention mechanism network,a hybrid network model suitable for text readability measurement is constructed,which replaces artificial automatic extraction of features by representation learning.Improves the efficiency and performance of text readability.(2)An English text readability metric based on hierarchical hybrid network model is proposed.The previous research and the hybrid network model used in this paper treat the whole document directly as a long sequence of word sequences.This way of processing makes the feature representation of the text lose the logical relationship of the text,the context information of the sentence.Related features.It is also considered that humans regard documents as consisting of sentence sequences,and sentence sequences are not consistent with the recognition of word sequences.In view of such deficiencies,this paper introduces the concept of hierarchical hybrid network model,constructs a hierarchical convolutional loop attention mechanism network model to measure the readability of English text,and proves that its model has good performance through experiments.(3)A loss function is designed to measure the readability of the text.Generally,the text readability metric based on deep learning is to treat the text readability metric as a text categorization task,and the superior performance of the cross entropy loss function in the classification task is difficult to overcome by other loss functions.However,because the text readability measure has a gradual relationship between the text reading difficulty level labels,there is a fundamental difference between the independent classification tasks and the general classification task labels.Based on this consideration,this paper uses cross entropy.The loss function is mainly based on the mean square error loss function,which constructs a main and auxiliary double loss form to jointly train our network model.It proves that the performance of the whole model is improved by experiments.