2003 baseline paper - Githubissues

ai-se / LDAClassification

Tuning LDA

https://github.com/amritbhanu/LDADE-package

0 stars 0 forks source link

2003 baseline paper #2

Closed amritbhanu closed 8 years ago

amritbhanu commented 8 years ago

According to Manny, this was the baseline paper - Pg 1012 which he talked about for feature reduction.

They evaluated using accuracy, and doc topic weight features for SVM classifier.
K=50 constant. Dataset used was Reuters-21578.

I ran similar experiment(k=50) on our manny mock datasets. Results are really bad. fscore of less than 10%. Even with tuning it had no effect. This could be because of highly unbalanced classes that even smote couldnt help.

amritbhanu commented 8 years ago

[ ] next step is to run it on Reuters dataset and see if tuning and not tuning helps or not.

timm commented 8 years ago

so you've got something to show to LN on thrusday. cool.

amritbhanu commented 8 years ago

there was a bug prof. Please ignore this issue. I am waiting for further results.