nTopic =
keyword
nNeed to track relevant words w.r.t. time
¡tf ´ cidf
(cumulative idf); corpus is a moving window
n
nFind three distributions of topics
¡Chatter: topics continuously discussed (e.g.,
alzheimers)
¡Spike: topic exhibiting a usage spike, then inactivity
(e.g., chibi)
¡Spiky Chatter: Topics (e.g., microsoft)
nOverlay of above two types (multiple spikes
possible)
nSpike removal possible with spike model