Font Size: a A A

Research On Text Sentiment Analysis Based On Support Vector Machine

Posted on:2015-07-12Degree:DoctorType:Dissertation
Country:ChinaCandidate:K X HanFull Text:PDF
GTID:1108330461983249Subject:Oil and Natural Gas Engineering
Abstract/Summary:PDF Full Text Request
The rapid development of the Internet, as the major platform for information release, delivery and impact, deeply influences all areas currently beyond the traditional media. The sentiment detection of Internet public opinion aims to grasp the social state and dynamic event, and has a great help for the decision of government, enterprises and agents. Aiming at the problems existing in the main text analysis method, in this paper, we mainly investigate the sentiment analysis of public opinion and focus on the study and improvement of the probabilistic latent semantic analysis, Fisher kernel function of SVM, Local multi-core learning and other related methods. Besides, the research results are applied to the text emotional tendencies judgment, providing a new idea for the analysis of emotional tendencies for network public opinions. The main work of this paper is as follows:1. The Fisher kernel method based on probabilistic latent semantic analysis is proposed. By means of this method, latent semantic information containing the latent semantic information can be used as the classification characteristicsit, the effect of classification for support vector machine (SVM) can be improved, and the problem of not considering the potential semantic characteristics in text sentiment analysis is solved.2. The parameter selection approach method of SVM based on Fisher discriminant analysis is proposed. Based on the study of Fisher discriminant analysis, for the problem of random initialization parameter in SVM, in the feature space, the parameter optimization is conducted combining the separability between classes of the sample. Furthermore, the problem of instability for experimental results in random initialization parameter is solved.3. Local multi-core learning algorithm is proposed. By means of this method, Local optimal kernel function can be selected by threshold model, which can be used to determine more effective sample characteristics. This algorithm is used for the choice of emotional key words in the text sentiment analysis, and the common problem of dimension disaster in the text is solved.4. According to the implied correlation between latent semantic analysis (LSA) and probabilistic latent semantic analysis (PLSA), the parameters initialization method of probabilistic latent semantic analysis is improved and used to provide a classification feature for text emotion. Three kinds of text sentiment analysis method based on support vector machine (SVM) are proposed, and the document is represented by text subject, which is a feature with high-level semantic information. The text features of probability and improved kernel function of support vector machine (SVM) are combined, thus to mine the emotional tendency in the text.5. Based on the above research, the precision and recall rate of three methods of text sentiment analysis were experimental verified using the Twitter data set. At the same time, the classification efficiency before and after the dimension reduction of feature word is contrasted. Based on the experiment results, the effect of sentiment analysis method is contrasted and confirmed. Finally, the application of the methods in this paper is carried out on the trend analysis of oil field technology, detecting its text emotion classification effect in the application.
Keywords/Search Tags:Support Vector Machines, Fisher Kernel, Local Multi-core Learning Algorithm, Probabilistic Latent Semantic Analysis, Text Sentiment Analysis
PDF Full Text Request
Related items