Research On Text Classification Method Based On Attention-Based Bi-GRU Model

In the era of big data,a large amount of text data has been generated by various terminal devices.Text classification technology,as the basis and key technology of natural language processing such as intelligent search and intelligent question and answer,has been a research hotspot in the interdisciplinary field.Since the text is unstructured data,the semantic information of the context of the text is neglected and the classification accuracy is not high when the text classification is solved by the traditional machine learning method.With the continuous development of deep learning technology,deep learning technology has been proved to have many advantages in extracting text features and improving the accuracy of classification.In the thesis,based on the analysis of GRU and Attention mechanism,the mixed deep learning model to solve the problem of text classification was deeply studied.The main research work of the thesis is as follows:(1)The thesis gives the mathematical definition of text classification and studies the general process of text classification,including text preprocessing,text representation,classifier design and performance evaluation.For the word vector representation,the Word2 Vec is studied and adopted,so that the word vector represents similar features with synonymous words,and avoids the difficulty of training neural network due to too high dimension.The thesis reviews some traditional text classification methods and points out their defects.(2)In the thesis,based on the research of the GRU and Attention mechanism,the thesis proposes a hybrid model of text classification——Attention-Based Bi-GRU.GRU solves the problem that traditional RNN often faces long-distance dependence when encoding sequence data.Bi-GRU comprehensively considers the context information of the text.The Attention mechanism can effectively highlight the key points,retain only the key information useful for the current task and discard the useless information.Experiments proved that the model improves the effect of text classification.(3)The Attention-Based Bi-GRU combines text feature extraction and softmax classifier.The text features extracted in training can be used for classification,and the results of classification will also act on the feature extraction process to achieve the effect of optimization.The method provides an end-to-end text feature extraction method,which does not rely on artificial feature engineering,and greatly saves human and financial resources.