Study On Text Sentiment Classification With Improved RBF Neural Network

Study on Text Sentiment Classification with Improved RBF neural networkBlog and twitter allow internet users to express their feelings about products or anything they have interests in. The internet directly presents personal feeling, including reviews on product or attitude towards some events. This sentimental information has great significance on internet users, enterprise and government.For personal, e-business make it convenient for users to buy the product they want, but before they make decisions, it is hard to get direct information about the product; so other consumers' reviews has reflect on them. There are too many information on internet, get useful information is getting harder and harder. To process this information is very useful for us. For enterprise, to get information about their product is important to them, especially how it feels in consumers' hands. Get the review information to improve the product, or to analysis the competitors' information to make improvement is important for the enterprise. With internet, the product review information is showed directly to them, and to process this information is really important. For government, they can get the individual's express especially about those political information through internet. How to process this information is good for the government to make better political decision and rules.The previous sentiment analysis algorithms are primary based on subject analysis, such as n-grams subject classification. Now vector space model (VSM) showed better result. It is easier to process the document data through vector space model (VSM) and to analysis the information through machine learning algorithms. This thesis analysis three different vector space model (VSM), the experiment result showed that the adverb and adjective based vector space model (VSM) outperform the vector space model based on keywords. Not only is the best classification result, but also smaller dimension.This thesis also presented a new RBF neural network, the AM+RBF neural network. Based on the vector space model data, the AM+RBF neural network showed better result than the Naive Bayes Classifier, increased the accuracy by 4.6%. and is not worse than Support Vector Machine (SVM). increased the accuracy by 1.9%.There is not much research on sentiment analysis based on neural network. The proposed AM-RBF neural network is a efficient algorithm especially in text sentiment classification. This thesis presented a new way in this research area.