Predicting Twitter Hashtags Popularity Level

This paper investigates the problem of predicting Twitter hashtags popularity level. A data set of more than 18 million tweets containing 748 thousand hashtags has been prepared by using Twitter's Streaming API. Early adoption properties including profile of tweet authors and adoption time series are used to predict a tag's later popularity level. The followers count and tweets count are two such characteristics related to adopters' profile. On the other hand, two types of frequency domain analyses are used to augment the simple mean and standard deviation characteristics of the adoption time series. Fourier transform (FT) spectrum and wavelet transform (WT) spectrum are considered in this study. Experimental results show that WT spectrum improves the prediction result of viral hashtags while FT spectrum does not.