Statistics of collective human behaviors observed in blog entries

Collective human behaviors are analyzed using the time series of word appearances in blogs. As expected, we confirm that the number of fluctuations is approximated by a Poisson distribution for very-low-frequency words. A non-trivial scaling roperty is confirmed for more-frequent words. We propose a simple model that shows that the fluctuations in the number of contributors is playing the central role in this non-Poissonian behavior.