search

bwbaugh / infertweet

Infer information from Tweets. Useful for human-centered computing tasks, such as sentiment analysis, location prediction, authorship profiling and more!
http://infertweet.bwbaugh.com/
Other
10 stars 1 forks source link

Try different labels for the hierarchical classifier levels #21

Open bwbaugh opened 11 years ago

bwbaugh commented 11 years ago

The intuitive approach for the hierarchical classifier (gh-4) is to have a subjectivity classifier first (objective vs. all), and if not objective then try positive vs. negative.

What about trying positive vs all, or negative vs all instead at the first level? In this particular case, it's probably best to use neutral vs all for the first level, but for other types of problems it might not necessarily be the best case.

Idea was originally recorded 3/16/13 at 21:25.

bwbaugh commented 11 years ago

The reason neutral might be best for the first level is because it is the majority class. The whole process of deciding what label should be at each level of the hierarchical classifier might be similar to the use of information gain in decision trees.

arita37 commented 8 years ago

Hello,

Did you get a chance to classify hiearchical classifier ?