| Summary: | 碩士 === 國立政治大學 === 資訊科學學系 === 103 === Nowadays, the use of social media is increasingly popular all over the world.
People can easily express their thoughts or receive information that they are
interested in via social media. Many studies have focused on exploring the
predictive power of the large amount of data generated from social media.
In this thesis, we address the problem of predicting the political preference
of social media users given the data of their past activities on Plurk and
evaluating our approach on the Taiwan 2012 presidential election. We first
collected Plurk messages posted six months before the election day. By building
predicting models based on a variety of contextual and behavioral features, we
find that predicting political preference of active users achieved up to 94.08%
classification accuracy. In the meanwhile, in order to extend the usability of our
work, we further use our models to analyze the change of user political
preference based on political events which happened before the election.
Identifying people who change their political preference frequently or stay
neutrally allows a candidate to design strategies to affect these people. All of the
political events are automatically selected by the popularity of political
keywords used in Plurk, and keywords can be extracted from daily political
news. In the end, we get 208 swing voters from 275 voters, who become the
main targets for enhancing the effectiveness of the campaign strategy.
|
|---|
