Research On Text Sentiment Classification Based On Improved Adaptive ISVM

In the background of the development of 5G and the Internet of Things,kinds of text,voice,and video user data are continuously generated,and it is necessary for the public to mine and analyze the inherent knowledge and commercial value from the data.The classification of text sentimentality is a research hotspot in the fields of machine learning and natural language processing.The research on the classification of sentiment orientation of text content plays an important role in the fields of product recommendation,government agency public opinion supervision,and judgment of the trend of social hot events.According to the characteristics of diverse network text content,complex structure and dynamic generation method,traditional SVM supervised machine learning algorithms cannot carry out large-scale data and incremental learning.The SVM method of incremental learning thinking is used to classify the emotional tendency of text content the study.Due to the existing SVM incremental learning algorithm,the selection of training samples in the learning process is unreasonable,and there is room for improvement in classification accuracy,training time and stability.Proposed.an improved adaptive ISVM algorithm,ND-ISVM algorithm.First,PCA technology is used to reduce the dimensionality of the data,and then the Mahalanobis distance and the geometric distance from the sample to the hyperplane are used to construct an adaptive boundary sample retention set for the original sample and the new sample.In the training process,the sample similarity coefficient is constantly adjusted to ensure the balance of sample classification.The performance comparison experiment between ND-ISVM algorithm and three other similar ISVM algorithms shows that the ND-ISVM algorithm has better performance in the performance analysis of accuracy,recall rate,F1 value,AUC stability and classification time.In the case of a slight improvement in classification accuracy,it can classify unbalanced data and maintain good model stability.The ND-ISVM algorithm is applied to the research of text sentiment tendency classification.The text content first uses the Word2vec++ algorithm to initialize the word vector,sentiment value weighted word vector,feature reduction,etc.for feature extraction preprocessing,and then use the ND-ISVM algorithm to Text sentiment tendencies are classified.Perform performance comparison experiments with other three similar ISVM algorithms.The results show that the classification accuracy and stability are better than the other three comparison algorithms under the premise that the training speed is not affected.The accuracy is maintained between 0.8 and 0.825,and the AUC value Up to about 80%.