Effective Clusterization of Political Tweets Using Kurtosis and Community Duration

Abstract

Exploration of voter opinions is important for policy making. While opinion polls have long played an important role in this process, big data analysis of social media, i.e. "social listening", is becoming important. This is because social listening involves the collection of a huge amount of data on opinions that are transmitted spontaneously by people in real time. The amount is so huge that the data needs to be aggregated and summarized. Graph theory is an effective way of aggregating into groups network structured data collected from social media such as Twitter. However, there are two challenges. One is to combine the groups, i.e. "communities", into clusters because the granularity of the community is too fine for understanding the big picture. The other is to distinguish insignificant clusters from those that contain relevant information. In this paper, we describe a method for community clustering that is based on kurtosis and duration in time series of each community.

DOI: 10.1109/SocialCom.2013.144

3 Figures and Tables

Cite this paper

@article{Itsuki2013EffectiveCO, title={Effective Clusterization of Political Tweets Using Kurtosis and Community Duration}, author={Hiroshi Itsuki and Hitoshi Matsubara and Kazuki Arita and Kazunari Omi}, journal={2013 International Conference on Social Computing}, year={2013}, pages={928-931} }