Design And Implementation Of Short Text Emotion Prediction System Based On Attention Parallel Double Convolution Model

The upgrading of mobile Internet and the rapid popularization of mobile intelligent terminals promote the networking of human life and lead to the generation of a large number of user comments.Through analysis of user comment information,understanding the emotional tendency of users is of great help for merchants to understand the market trend and improve the quality of goods.Therefore,a large number of researchers have studied the task of text emotion analysis and achieved a lot of results.In these studies,emotional analysis is essentially the learning of text information,which is best represented by CNN model and LSTM model.CNN model is good at extracting local features of text,while LSTM model is good at extracting semantic information of text.Researchers often use the combination of the two models to fully extract text features.However,LSTM has serious resource consumption and model training time.So,APDCNN model is proposed and the idea of matrix decomposition is used to construct emotion prediction system.The APDCNN model is mainly composed of two parts,1D-CNN and 2D-CNN.1D-CNN is a model which is more suitable for extracting semantic features of text obtained by traditional CNN after a series of improvements and adding attention mechanism.2D-CNN is a traditional CNN model with two-dimensional convolution.The two models extract text features in parallel,and then form APDCNN model.In the paper,APDCNN model is used to analysis the comments to get the emotional tendencies.Then emotions tendency value matrix is built,which is decomposited to user characteristic matrix and characteristic matrix by matrix decomposition and deep neural network.After that,the two matrices are used to predict the user emotions tendency value as the emotion prediction system completed.Through comparative experiments with CNN model,LSTM model and lstm-cnn model,it is found that the classification accuracy and F1 value of APDCNN model reach about 90%,15% higher than the CNN model,13% higher than the LSTM model,and 3% higher than the lstm-cnn model under the same condition.At the same time,the training time of APDCNN model under the same condition is one twelfth of that of lstm-cnn model,which is far less than that of lstm-cnn model.The affective prediction results of the prediction system are also generally consistent with the actual results.