Abstract: How to discriminate accurately the microblog topic position is one of the highlights in the short essay mining. This paper proposes a method based on the text classification with correlation of subject, which can discriminate users for the topic who is to support or oppose. The correlation of subject often leads to the text that have greatly different features. The method first obtain the topic keywords by extraction technology and mutual information, then classify the text to conversation corpus with the correlation of subject, at last adopt different method to analyze the comprehensive microblog topic position. The experimental results show that the method of correlated adopting machine learning and the uncorrelated adopting dictionary can greatly improve the discrimination accuracy. On this basis, we construct a model, can be used for the relevant government departments to monitor the Internet public opinion and business evaluate the products market, etc.

Key words: microblog, topic, position, correlation, naive-bayes