In the following, we plot the top 5 hashtag series (dot-dashed) overlayed with the first principal components scaled with corresponding coefficients. Series are de-centered. This validates the previous observation that principal components filter out the singleton spikes.