Nowadays, many people express their opinions on the Internet, including opinions about political issues. Not only governments need to know about their citizens’ opinions, but also most election campaigns and candidates are eager to know the voice of voters before the election so they can adjust their election strategy promptly. Over the years, election–related organizations have conducted polls forecasting, and the election results are mostly consistent with the prediction. However, in the Taiwan local elections held on 29 November 2014, the election results of many cities showed significant deviation from predictions. For example, the pre-election polls of the mayoral candidate of the Taoyuan city, Chih-Yang Wu, were 57 percent favorable, and those of his opponent, Wen-tsan Cheng, were only 25% favorable. However, the polls’ laggard turned out to be the winner. Evidently, simply relying on the traditional polling method doesn’t seem enough to provide an adequate forecast. With the advent of social media, such as Facebook, Plurk and Twitter, users can express their opinions anytime and anywhere. In general, young people are the main users of the Internet. As shown in Table 1, 24% of the Internet access population in Taiwan are between the ages of 15 and 24, 29% of the Internet access population are between the ages of 25 and 34. Hence, using Web reviews for election estimation mainly reflects the opinions of young people. **Insert Table 1 here** There are many poll centers in Taiwan that generally collect data using traditional methods, such as via questionnaire and interview, which are expensive and time-consuming. In recent years, there has been increasing interest in the use of data mining to translate massive amounts of social media data into useful information. The new method of using Web mining for polling is not only fast and cost-saving, but also can be used to investigate the opinions of populations that have not been reached before; it is therefore expected to replace or combine with traditional polling methods. In view of the problems faced by traditional polling, such as inadequate data for quota design and poor response rates for phone polls, this study utilized the opinions posted on social media to gauge the degree of support for two mayoral candidates of Taichung city. However, to effectively and quickly extract useful information from tons of Web data is an important issue. In particular, there are many problems in conducting opinion analysis on comments in Chinese. Hence, before performing election prediction based on Chinese social media sites, the opinion extraction rules of Chinese texts should be established first. The remainder of this paper is organized as follows. Section 2 introduces the development of social media mining in recent years with special focus on its application to election prediction. The problems of conducting semantic analysis on Chinese social media are also presented in this section. Section 3 depicts the framework of this study. The details of constructing opinion phrase extraction rules for identifying the opinion words associated with the attribute words are described. This section also introduces the six municipal governancerelated topics. Section 4 presents the results of the scores of the six topics of each candidate, based on which the election prediction is performed. Finally, Section 5 presents conclusions and future work.