----------------------------------------------------------------------------
----------------------------------------------------------------------------
----------------------------------------------------------------------------

# KDD step 4: interpretation


golden standard

• --> annotated test set
• 
  

10-fold cross validation

• taking 1000 tweets
• training 800 tweets
• test 100 tweets
• val 100 tweets
• 
  

compare to baseline scores

• - frequency-baseline
  • you expect 80% of the tweets to be neutral(?)
• - informative baseline
  • i have a 60% chance that it will rain tomorrow
  • --> your result need to be higher
  • otherwise --> why do all the work?

from : CLiPS ­ Guy de Pauw, Pattern workshop — Cqrrelations, January 2015


----------------------------------------------------------------------------