-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Chris McKenzie edited this page Jun 17, 2015
·
2 revisions
This is a discussion on methodologies to classify small phrases in a fairly-context-free way on the dimension of appropriateness and tact.
The first approach consisted of:
- stop word removal & input cleansing
- n-gram analysis with manual weights attributed to key(word/phrases).
In a vocabulary based approach, the nuances of a semantic abrasiveness is lost.
A rough sketch of what's needed.
There should be a multi-pronged approach to aggregate classification. Some of the keyword systems that are used here are viterbi and perhaps this paper. More generally, we are talking about ensemble learning.