As new technologies advances, internet become more popular. More and more person states their opinion on internet. In democratic society, people have suffrage and freedom of speech. People always share their opinions about policy on the internet. To know the opinions of people, we must employ a lot of employees to make phone-based poll in the past. Nowadays, we can crawl and download the articles easily on the internet and use the text mining techniques to deal with political issues. Then, we can estimate the sentiment orientations of political articles and show the political orientations of internet users. This paper uses semantic orientation from PMI method and information gain method to add sentimental terms in sentimental dictionary. We make comparison between TF-IDF variable, sentimental variable and combined variable models in the classification of sentiment. We also build the model of sentiment analysis and develop an internet public opinion index to estimate the degree of satisfaction to administer of Taipei City Government of Taipei’s citizen. With the best accuracy and excellent stability, Support Vector Machine is the best choice for us to do the sentimental classification. If the topic of training data is similar to the topic of the testing data, the testing accuracy will be higher. Do text mining analysis for internet texts is helpful for us to analyze internet public opinions.