analysis_ideas

STATISTICS:

query language: statistic type & subset to look at
ex, statistic type could be most used words, subset could be wednesdays over last month

STATISTIC TYPES:
1) m most used word n-grams
2) m least used word n-grams
3) sentiment distribution (e.g. this many appeared very happy, this many appeared not that happy, etc)
4) number of appearances of a particular n-gram

SUBSETS (can be intersected, i.e. combined):
1) day of week
2) grade (more generally, answer to a special question)
3) answers to question Y that start with string X 
4) answers to question Y that include string X


I definitely think an interactive console is the best way to do this
It would be weird and burdensome to do journos analyze x, journos analyze y, etc.



GENERATING RANDOM ENTRY:
Markov chain style, sampling from the distribution on entries in a particular subset (see above for subsets)