Research On Improved Dense Net Algorithm In Short Text Classification

With the in-depth development of artificial intelligence technology,the number of short text messages has gradually shown a blowout trend.How to effectively classify and manage these massive text messages has become one of the current hot research topics.The traditional classification method not only requires a lot of manual design cost,but also cannot effectively classify the category of the new field.Unlike this,the deep learning method uses hierarchical feature extraction to automatically find out the important features required by the classification problem,and efficiently process the text data.Therefore,this paper studies the short text classification algorithm based on deep learning,and proposes a corresponding high-efficiency classification algorithms.The main innovations of the thesis are as follows:1.In view of the defect of feature loss in the training process of short text classification algorithm,this paper proposes an attention pool-based convolutional neural network(Attention Pooling CNN,AP-CNN)text classification algorithm.The idea of the algorithm is: by means of attention mechanism,the convolutional layer vectors are weighted linear combination,so as to obtain the best features in dimension and retain more valuable text classification information.Simulation experiments prove that AP-CNN short text classification algorithm has certain advantages in accuracy.2.In order to further improve the accuracy of short text classification,this paper considers deep-level text information features,and proposes a Dense Net short text classification algorithm based on the attention mechanism(Att-Dense Net).The idea of the algorithm is: a multi-level attention mechanism is introduced into the dense layer of Dense Net to optimize the global representation of the sequence in the text,so that the algorithm can automatically select and focus on more important text features,and further extract deeper information in the text information.The simulation experiment results show that the Att-Dense Net short text classification algorithm can effectively alleviate the problem of network degradation on the basis of further improving the accuracy of text classification.3.Aiming at the problem that the insufficient data amount leads to overfitting phenomenon in the training process of neural network,which will reduce the accuracy of the algorithm,this paper proposes an Att-Dense Net short text classification algorithm based on Easy Data Augmentation(EDA),namely EAtt-Dense Net.The idea of the algorithm is to use the four operations of the EDA Algorithm,namely synonym replacement,random deletion,random swap and random insertion,to enhance the data set of text classification and further verify the Att-Dense Net network model.The simulation experiment results show that the short text classification algorithm of EAtt-Dense Net successfully reduces the number of occurrences of overfitting,and the accuracy is better.