The Research On Multi-feature Fusion Method For Sentiment Analysis Of Chinese Microblogging

In recent years,with the rapid growth of microblogging registered users,the user data accumulated by the microblogging platform has increased geometrically.Microblogging sentiment analysis is a hot research area,which investigates into effectively dicover valuable information from these mass data.Microblogging sentiment analysis has been widely applied in the field of public opinion monitoring,hotspot tracking,and user satisfaction survey.However,due to the fact that the microblog text has less information,the format is not standardized,the spoken language is serious,and it contains a large amount of noise compared with the traditional long text,the performances of the microblog emotion classification methods are less than satisfactory.In addition,due to the particularity of the Chinese language,the accuracy of the Chinese Weibo's sentiment analysis is more unsatisfactory.To improve the performance of the microblog emotion classification,this thesis proposes an approach of merging multiple features to perform sentiment analysis,extracting score features based on sentiment lexicon,probability characteristics based on machine learning,and word vector feature fusion based on deep learning to perform sentiment analysis.The main contributions of the thesis are as the following:1.Investigate into the principles and algorithms of the two sentiment analysis methods basing on the emotion dictionary features and machine learning features.And their advantages and disadvantages are also compared.2.The method of expanding the sentiment dictionary based on SO-PMI algorithm and emotional symbol dictionary is designed.The extended dictionary is combined with the rule template to extract the emotion score feature of each text.Experiments show that using the new emotion score feature classification effect is better than traditional emotion dictionary features.3.A new microblog text feature extraction method based on machine learning is designed.Using the idea of ensemble learning,the probabilistic output results of KNN,LR,SVM and nearest neighbor classifiers are integrated to construct feature vectors as features.Experiments show that the use of new probability features based on machine learning for emotional analysis is better.Characterization ability is stronger4.Word2 vec algorithm is used to extract the word vector features containing contextual semantics as a supplementary feature.At the same time,emotion classification experiments are performed by combining the emotion score feature and machine learning feature probability.Experiments show that the method of merging multiple features for sentiment analysis proposed in this paper can achieve feature complementarity and achieve better emotional classification effect.