Sentiment Inyensity Analysis Of Chinese Douban Film Criticism Based On Machine Learning

Text sentiment classification is a hot topic in the field of Natural Language Processing.Text sentiment classification is a research topic,which involves Linguistics,physiology,computer science,Natural Language Processing.Based on the short film reviews of Chinese Douban,the author uses two important text mining methods,machine learning and deep learning to study classification of emotion and degree of emotion,propose a structural model which combines machine learning and deep learning and improves performance of the structural model and the effect of emotion classification.The main contents and innovations of this paper are as follows:1.The background and significance of text classification,summary of current research status of Chinese film reviews and a describetion of unsolved technology and theory problems.The background and significance of deep neural network,which explains the structure of deep learning method and other common methods derived from it and summarized the status of deep learning in text mining.2.Based on the opinion sentences extracted in this paper,the classification effect of emotion intensity between the traditional machine learning model and the deep neural network machine learning model is compared,and a classification model combining the two methods is proposed.3.Built three categories of emotional knowledge base,which including and Expanding the field of unrelated emotion words,related emotional words and network language.A comparison of text feature and emotional feature,learning methods based on rules and based on machine learning and the effect of subjectivesentence extraction between different algorithm models.4.Proposed a classification method based on hierarchy.Divided the problem of emotion classification into two layers,the first layer applies the SVM model to three classifications of emotional tendencies and the second level applies the CRFs model to the data base which distinguished between positive and negative emotional tendencies into five classifications of emotional strength.In order to reduce the feature masking rate,mark redundancy and calculation scale of two layers.5.Based on the adtantage that CNNs can reduce network model complexity,the number of weights and extract multidimensional features,proposed a method which uses convolution neural network to extract local features and put emotinal labels on CRFs algorithm.The experimental analysis shows that the F value of SVM+CRFs and CNNs is 4.5% and 3.6% higher than that of hierarchical ones.