Research And Appliccation On Text Sentiment Classification Based Nn Deep Learning

With the rapid development of various social networks and shopping platforms,the popularity of smartphones and the increasing generation of text data have promoted the research and development of text sentiment analysis.Fast-food life hinders communication between people and creates greater generation gaps and communication barriers between teachers and students.Students with psychological problems have increased year by year.Students are used to doing some online speech and expressing their feelings on WeChat,QQ or Weibo.The teacher can analyze the speeches sent by the students to better understand the inner world and emotional tendencies of the students,find out some problems of the students in time,give timely guidance,and avoid the extreme development of the problems.Traditional machine learning methods are not able to meet the needs of modern people because of the manual selection and construction of features.This paper studies the sentiment analysis at home and abroad,and combines the characteristics of Chinese to propose the "Research and application of Chinese text sentiment classification based on deep learning".This thesis deeply studies and analyzes Convolutional Neural Networks and Cyclic Neural Networks,and deeply studies and analyzes classical neural network models and algorithms,such as ADAM algorithm,Batch Gradient Descent Algorithm,Stochastic Gradient Descent Algorithm,and Small Batch Gradient DescentAlgorithm.Gradient descent algorithm,Momentum Optimization Algorithm,etc.On the basis of the above research,the research results are as follows:1.There is still room to improve the performance of the model.Inspired by common methods in image processing field,the original character-level convolution neural network is combined with local response normalization(LRN)layer to establish Char-CNN-LRN model.The LRN normalization layer is used to optimize the classification algorithms,such as student sentiment analysis and so on.Compared with the basic model Char-CNN method,the experimental results in Chapter 4 show that the performance of the normalized character-level convolutional neural network classification model has been improved to a certain extent.2.In view of the fact that the weight W and offset B in convolution algorithm are generated by random functions,which may affect the accuracy of the model and even the convergence efficiency,a parallel convolution processing method is proposed.On the basis of the previous research,a normalized character level parallel convolution neural network model classification model(Char-DuCNN-LRN)is established.All convolution computations are carried out in parallel,and then the results are weighted sum,so as to achieve the goal of model optimization.From the experimental results,it can be seen that the convergence rate of the model has increased significantly.The proposed Char-DuCNN-LRN model is compared with other popular models such as Linar model and FastText.The results show that the model is effective and superior.3.This paper applies the proposed normalized character-level parallel convolution neural network model classification algorithm to student sentiment analysis,and designs a student sentiment analysis system,which can realize student sentiment analysis,corpus query and background database operation and maintenance.The test results show that the expected results have been achieved.