Based On The Deep Learning Model Of Semantic Analysis And Processing Of Emotional

With the rapid development of the Internet,resulting in a large number of views and opinions contained in the data,which makes the problem of network information overload has become increasingly prominent,but these information also has an invaluable value,so the natural language processing has become a very popular field.The emotional analysis of text data is one of the main contents of the research in this field.In this paper,the text semantic emotional analysis as the main object of study,for the field of current Chinese emotional analysis,the basic concepts of text analysis and related theories and research status in China and abroad are introduced and discussed,the following research were conducted:(1)The influence of the use of different stop words list on the classification of emotional sentiment is studied.According to the traditional stop words list,this paper constructs a stop words list,which is specially used for emotional analysis.After the text is processed by different stop words list,the word frequency weight feature is extracted by using TF-IDF,and the SVM classifier respectively,for the feature selection and text orientation classification experiments.Experiments show that more emotional information is included in the text feature candidate set after using the stop words list that are devoted to emotional analysis,and the highest accuracy rate of the positive text classification is 81.94%.(2)Carefully studied the impact of the length of time on the voting mechanism,and set the corresponding voting standards.For the data processing of eight different categories of product reviews in the online shopping platform,we combine the features obtained from the statistical methods and the features obtained from the emotional dictionary method.Finally,the linear regression model is used to predict the review is useful or useless.Different product categories predict different results.The longer the existence of the review,the better the results of the prediction,so adding the time characteristics of the reviews vote can make the results of the emotional analysis more accurate and reasonable.(3)Research deep learning feature extraction methods.In this paper,the Word2vec tool based on deep learning is used to obtain the low dimensional word vector containing deep semantic information by calculating the similarity between words.In the study,the average of Word2vec training words was used as the characteristic,and the characteristics of the classifier based on the statistical method and the emotional dictionary method were used to make the input of the classifier.Finally,the SVM multi-splitter was used to classify the comments text.Experiments show that this method solves the problem of the traditional feature set can't reflect the text semantics,and improves the classification accuracy of SVM classifier.(4)In order to further improve the classification accuracy of samples near the interface,an improved SVM multi-classification algorithm is proposed.Using the classifier support vector as a standard,the sample identified as a certain class is judged according to the set threshold whether its position is near the interface.If the sample may be a false positive sample,the sample is judged by the second classification of the KNN algorithm category.Experiments show that in the four-class experiment,the classification accuracy can reach 85%and the F value is higher than the average classification of SVM 4 percentage points.Therefore,the SVM multi-classifier algorithm combined with KNN algorithm can effectively improve the classification accuracy.