E12-1035 |
guessing strategies . We now turn to
|
macro-averaging
|
across multiple classifiers or
|
C04-1137 |
in the test set . The choice of
|
macro-averaging
|
over micro-averaging does not
|
D13-1070 |
cision ) than summaries B . When
|
macro-averaging
|
, the Recall score of summaries
|
P11-2044 |
aligned token pairs per problem and
|
macro-averaging
|
over all problems . The results
|
P07-1097 |
distinction between micro - and
|
macro-averaging
|
often seen in classification
|
C02-1103 |
the micro-averaging method and
|
macro-averaging
|
method . 2.2 Experimental Results
|
P07-1097 |
distinction similar to micro - and
|
macro-averaging
|
in the context of classification
|
D11-1072 |
called micro-averaging , the latter
|
macro-averaging
|
. As we use a knowledge base
|
D12-1035 |
measures by both micro-averaging and
|
macro-averaging
|
. Micro-averaging aggregates
|
E12-1035 |
distribution of weights . Commonly
|
macro-averaging
|
averages across classes as average
|
C04-1070 |
then compute the F1 score . In
|
macro-averaging
|
, the F1 score is computed for
|
P07-1097 |
distinction between micro - and
|
macro-averaging
|
discussed in the context of text
|
P13-1141 |
precision/recall/f - measure . For
|
macro-averaging
|
, we compute a single confusion
|
I05-3015 |
categories . The way is called
|
macro-averaging
|
method . For evaluating performance
|
E12-1035 |
controversy over these averages , and
|
macro-averaging
|
in general , relates to one of
|
N12-2006 |
sentence in the treebank ( termed
|
macro-averaging
|
, in contrast to microaveraging
|
P05-1062 |
the basic evaluation metrics and
|
macro-averaging
|
strategy was used to calculate
|
D08-1114 |
row " average " was obtained by
|
macro-averaging
|
( average of av - erages ) .
|
N12-1021 |
and Flach , 2011 ) . However ,
|
macro-averaging
|
better reflects performance on
|
P13-1160 |
the least important , so we use
|
macro-averaging
|
over labels and then average
|