Results on R2001(Region part) data (228 categories in
training set vs 296 categories in test set)
5.1 KNN on 2001r
The
graph tuning feature selection number (10000 is good for both
micro and macro avg. performance)
The
graph tuning fbr value (0.5 is good for both micro avg.
performance and 0.1 for macro avg. performance)
The
graph tuning knn_k value (10 is good for micro avg. performance
and 100 is OK for macro avg. performance)
micro
avg. F1 Result and macro
avg. F1 Result
5.2 Rocchio on 2001r
The
graph tuning feature selection number (10000 is good for both
micro and macro avg. performance)
The
graph tuning beta value (-2 is good for micro and -2 for macro
avg. performance)
The graph
tuning fbr number (0.5 is good for both micro and macro avg.
performance)
micro
AND macro avg. F1 Result
5.3 NB on 2001r
micro
avg. F1 Result and macro
avg. F1 Result
(For macro avg. result, 6000 top features are
used. fbr=0.3. For micro avg. result, 2000 top features are
used. fbr=0.1.)
5.4 SVM on 2001r
The
graph tuning feature selection number (feature selection is not
necessary)
The graph tuning
fbr value (0.2 is good for micro and macro)
micro
AND macro avg. F1 Result
Conclusion: