Research On Convolutional Neural Network Text Classification Model Based On Attention Mechanism

With the rapid development of Internet technology,a large amount of text data is generated every day.As the basis of natural language processing and network information mining,text classification plays an important role in text information processing.Manual text classification methods and traditional machine learning text classification methods are difficult to meet the current requirements for text classification efficiency and accuracy.With the application of deep learning in natural language,the deep learning method provides a new solution to the text classification.Based on the research of text vector representation technology and deep learning model CNN principle,this paper makes an in-depth study on the use of CNN model to text classification.Aiming at the problem that the data sparse coding dimension is high and the carrying information is small in the text classification,the distributed word vector representation method is used to map the text data into the low-dimensional vector space to obtain the grammatical semantic rich word vector representation.For the sample noise problem generated by using the pre-trained word vector,the self-training word vector is used to replace the word outside the bag,which effectively reduces the classification sample noise.For the problem of CNN is suitable for extracting local features and it is difficult to extract non-local relations and semantic information between text features,and the CNN has been improved reasonably.The Attention CNN structure is constructed by combining CNN and self Attention mechanism.The common convolution operation is used to obtain the local state of the feature,and the Attention mechanism acquires the global state of the feature,which provides guarantee for extracting the important features of the statement level.Based on the above Attention CNN structure,an A-CNN text classification model for multitasking is constructed,and the model parameters are pre-trained using the supervised method.Finally,the effectiveness of Attention CNN structure is verified by experiments,and the A-CNN model is used to conduct experiments and comparative analysis on task datasets such as sentiment analysis,problem classification and question answer selection.The results show that the highest accuracy of A-CNN model is 1.9%,4.3% and 2.1% higher than that of the model in the three text categorization tasks mentioned above.A-CNN model is of better precision and generality in text classification tasks.