An Approach of Semi-automatic Public Sentiment Analysis for Opinion and District

The contents generated by netizens on the Web can reflect public sentiments to a great extent, so analyzing these contents is very useful for government agencies in guiding their public information, propaganda programs, and decision support. Because of the civilization diversity and economy difference, the netizens inhabiting or employing in different districts may have the different sentiments for the same topic or event. Analyzing the sentiment difference of different districts will help government agencies make more pertinent decision. However, current researches in this domain have less considered the opinion distribution on different districts. In this paper, we propose an approach of semi-automatic public sentiment analysis for opinion and district, which includes automatic data acquiring, sentiment modeling, opinion clustering, and district clustering, and manual threshold setting and result analysis. In detail, on the one hand, we group public sentiment into some opinion clusters by means of clustering technique. On the other hand, based on the opinion clusters, we further partition every opinion cluster on district into district opinion and analyze the result. Experiment results in Tencent comments show the feasibility and validity of our approach.