Search

With the rapid development of new network media, the quandty of oniine reviews has a tendency of explosive growth. Tradkional manual methods for opinion mining have some problems when deaiing wtth tremendous oniine texts, such as low efticiency, fuzzy ciassitication boundary, and iiirited domain-adapion ablitty. In order to solve the above proWems, we improve the tradtional latent Dirichlet allocation (LDA) model, and propose a LDA topic model based on domain identification for o-pinion mining of online reviews. Firstly, a domain layer is added to ihe standard LDA model lo sample the domain lags of ihe document, and field parameters are utilized lo solve the LDA model. Secondly, given the sentimental connection between sentences9 we insert a sentiment layer between the topic layer and word layer. Sentimental transition variable is introduced lo denote related characters, which can increase the accuracy of senlimenl polarity analysis. Experimental results verify ihe validity of ihe pro-posed model and theory. [ABSTRACT FROM AUTHOR]

How to automatically extract valuable opinion information from enormous texts has become an important technical challenge. Currently, most opinion mining methods are based on the assumption that topics are independent of each other. However, there are complicated inherent relationships between topics. In order to solve the above problems, based on standard CTM model, the paper proposes a hybrid correlated topic model that mixes topic with sentiment to perform opinion mining. Considering the topic correlation of documents, opinion characters and potential sentiment tendency are analyzed. Based on these results, sentiment polarity of the whole review and each topic are obtained. The simulation results verify the validity of the proposed model. R language is also used to visualize the experimental results. [ABSTRACT FROM AUTHOR]

How to automatically extract related information from enormous texts has become a huge challenge. As an efficient way to solve this problem, text classification has attracted much attention, in which text representation is a critical factor to affect classification results. The correlated topic model can implement text representation, which can correctly reflect the correlation between topics under the case to remain the integrity of information. Based on this model, we optimize feature selection and the number of topics, and determine the number of topics with perplexity and log-likelihood function. We adopt the principal component analysis algorithm based on mutual information to optimize feature selection, which can reduce data dimension and the redundancy of text features. The R language is used to visualize the experimental results. [ABSTRACT FROM AUTHOR]

Many current methods of opinion mining are coarse-grained, which are practically prob-lematic due to insufficient feedback information. To address these problems, we propose a novel topic and sentiment joint maximum entropy LDA model in this paper for fine-grained opinion mining. Considering semantic and location information of words, a maximum entropy component is first added to the traditional LDA model to distinguish background words, aspect words and opinion words. Both the local extraction and global extraction of these words are further realized. Secondly, a sentiment layer is inserted between a topic layer and a word layer to perform fine-grained opinion mining on word or phrase level. Transition variable is introduced to deal with sentiment dependency. The sentiment polarity of the whole review and each topic are simultaneously acquired. Experimental results demonstrate the validity of the proposed model and theory. [ABSTRACT FROM AUTHOR]