No abstract

Social Media become more and more popular and are also heavily used for communication about many different events in society. There is a trend in research studies to use Social Media data for predictions, especially in the political domain, while it is unclear which Social Media platforms are suitable, and if so, to which degree. Most studies focus on a single platform only. Using the 2012 U.S. Republican preelections as a popular example for a political use-case, this work tries to compare different Social Media platforms under different aspects, in order to get a first idea about their suitabilities, advantages and weaknesses in comparison. We monitored the seven major candidates by collecting publicly available data from blogs, Facebook, Twitter and YouTube. We investigate the potential of this Social Media data to be used as a predictor of the real world performance of these candidates. Our relatively simple approach shows a good correlation to the 2012 primary results as well as to public opinion polls regarding this election process. We see significant differences between the platforms and single anomalies demonstrate how fragile these methods really are. In conclusion, it is apparent that a critical selection and interpretation in this specific field is very crucial.

When monitoring blog articles for the tracking of a certain personality or product, the automatic identification of topic clusters is of high interest. Clustering by textual content is a popular method to accomplish this. In this paper we investigate how links between individual blog articles can be used to support this clustering with another dimension of information. Given the existing component structure of these networks, we focus on the extension with links based on shared social media resources. We show that the component structure extended in this way is of very high use for supporting textual clustering algorithms, and may be used for a new type of hybrid clustering algorithms in the future.