comment classifier

similar to Google Junk Mail filter

Data set

1. 10,000 comments

tag manually as:

1. low-IQ rant
2. endorsement
3. valid criticism
4. useful insight
5. sincere question

action

1. low-IQ rant - delete and block user
2. endorsement - keep hidden (read when feeling blue)
3. valid criticism - keep, read and send automatic thank you
4. useful insight - keep, read and send automatic thank you
5. sincere question - keep, read and send manual explanatory answer

Procedural algorithm

1. measure IQ
   1. all caps
   2. exclamation points
   3. vocabulary
   4. grammar
2. measure positive or negative, using vocabulary
3. identify questions, using grammar and punctuation
4. ?

Neural net

1. training using the tagged dataset