• No results found

Correcting Category Errors in Text Classification

N/A
N/A
Protected

Academic year: 2020

Share "Correcting Category Errors in Text Classification"

Copied!
7
0
0

Loading.... (view fulltext now)

Full text

(1)

!

"

#

$

% &'

%())*

+

! Æ

" Æ

!

# $%%%& !

'()

#! )

*++*& , -

Æ '

!.'/

!

0 !

.'/ 1.2 3

42

5

!

#5!*+++&

5!

67

0 8

"

!

!

Æ

.

'()#(! $%%9&

'()

:

:

7 .

(2)
(3)

Correcting Annotation Errors

Training Samples

SVM

NB

Loss Function

Judgements using Threshold Value

Final results

Support Vectors

Training samples - Support Vectors

Error Candidates

Training samples - Error Candidates

(4)

1.2 $%%B 1#12($& >

'()

$

'() !

#9& .

!

1.2 # $%%B&

C+*+D !

)' /

$%%E#) $%%9& .

. $??E$

.

$

6 7 $

)

#F

& #F &

#&

67

. B*D+%

2#)$%%D&

!

* #&

6'7

657

#&

#&

7

#&

8

*

> *

+C

+C

$ +*+ ! ! +*+ "

!" # #

#$ # # #% & #'" ('

$#$ "

!#+!+9&

5!

+?*+ +DB+

.

" * +$

$ ! +$E 62 7

62 7

.

0

20

40

60

80

100

1

2

3

4

5

6

7

8

9

# of corrected samples

Round

corrected by the system

corrected by a human

" *0 G

"#

.

+

$ ! +$%

6,7 6'()7

C

, '()

, "#

< $#"$%D9& 6,7

6F 27

" ,

*CH '()

$DH ,

++$#I $%%%&

$$%

$%%B 1#1 *+++&

?+BD%$*+ $%%B

)'!#&"' #

!#&' '#*$

(5)

(## (#

+&# ,# ,# -# .,#

# *.//0*.//, *.//, 0*./-/, *./-/,0*.// *.//0*.//

*0 F #1.2&

1!" 2" 3

" , " "-4/, " "-4/,

". " "-/ " "-/

", , " "-,,/ " "4/

"- " "-,,/ " "4/

"4 , " ",/ " "-,/

" - " ",/ " "-/

" . " "-./. " "-,,/.

" 4 -, " ",/4 " "-4/4

" 4 , " "-.,/- "

"-/-" ,. .- " ,/ " -4/

4 , ". "-/-, ". "4,4,/-,

C0 2 #1.2&

56#%

(## 7# 8

6# "4, "44 "4

#$ "4, "4-

"4-19:#%

(## 7# 8

6# "4, "4.

"4,-#$ "4.4 "-

$% $%%D

- $*B

*9 CC EC $

E . $+*

. C*+%C9

#' $%%9&

C*$

!

9 !

1.2

:!#+!+9& 5

9

+?$% +D9E

1.2

+?*+ +DB+

!

> $%%B 1

*9C? $9++++

#$B%H&

$E+++

?+BD%$

F

#+?$%&

B

> C 9 B

62#22 &7

9$% *?B*

D+9 *?B*

+

$ ! ++B

D+?*H

622 7

622 7

I

$

%# # :

#I

$%%%&

"#

D

,

(6)

(## (#

+&# # # # 444#

# *.4/,/0*.4/./ *.4/./ 0*.4//, *.4//,0*.4// *.4//0*.-/,/.

90 F #1$%%B&

1!" 2" 3

" 44 ., " "-..-/4 " "4-4,/4

". .4 , " "-,./- "

"4../-", .., .4, " "-4/-. " "4-./-.

"- , " "-44-/- "

"4,/-"4 44 " "-44,/-,. " "4.4/-.,

" ., , " "---4/, " "-4/,

" 4. ,- " "-./, " "-./

" 4 4 ", ",444/, "- "--,/

" -. - " 4/.4 " "-4./.4

" .444 ., ", ",/4, ", "--/4,

-4 , "4 /,4 "4 -/,4

D ,

$DH'()$$H ,

:

++$

D0 2 #1$%%B&

56#%

(## 7# 8

6# "- "4- "4.4

#$ "-, "4,.

"-19:#%

(## 7# 8

6# "- "-

"-#$ "-. "-

?

"

'()

6 7 "

BEH

?

62$9*7

659$*7 6 "7 >

#F2*+++&

'()

9++++ #$+* &

BCE;A-

?0 , %())*!

!" " # $%&## $%'(

) ( $%#*( $%+$&$%$&

,- * $%#'' $%+$$%$*.

/

0 ( $%' & $%&+

",,1( $%*'& $%*(.2$%$.

0) $%.(& $%..'2$%$ +

>( 2* ;, 1 ) 5Æ

!

: 0 ,

'()."

1.2

$%%B 1

" !

Æ

#J*++C&

"

! !

!

(7)

( 6& ;&

3<$$#$$

9#'#$#&=" 8& # 2>< #?

@1;58#; A$#&B# #<" B# 1#<7#

3<$$<$

7(710;## #A0$!$$$#<" ; 3#

C$#!#<&$#0;" ; 2

C<$$#$$

($0;" 2>< #? 3#

4 :A'#$#<$#<$" 2>< #? 3#

@

- 3##&##$" 7#/8<

, C$#!#<&$#0;" 8& # 2

1"5

2 3 " >

'

#$

' 15 ' I ' $%%%

,

>

!"

C?KE9

' F A 2 *+++ A

2 . 2 >

#$ %

% &

%& *9BK*BC

55! *+++ F5 2

F >

'

(

$E?K$9C

, " $%D9 >0

2

> 2 "

>

*EBK*B9

) $%%9 )*

2

I ) 3 I

I A > >

$%%D " "

>' 1

>'>'1%D++D

3 )2 $%%% ) 2

) )

5) > %& !

+)) ,- .

!I) *++* F

5 2 4 ' (

) > ()

D+%KD$9

1 *+++ % !(

/())'01#0())201()%

#000((0$ 3 ! (

1

1 3)2 *++$

'5

51 >

(1

EE$KEE?

A ' $%%9 >

' ;

>

/ ! A >

1 ! $%%B F 1.2

F 22#

/& > )'($/ ?%K

%B

( (! $%%9

" ' I!

/ . 2 .! $%%? )2

' ( ) > %

%)10*

II L $%%% 15

2- ) >

## %

% &

%& E*KE%

/ J 1 / I I ; A

*++C ) 10

'()

' 2- >

#0

References

Related documents

Analyzing the results obtained from the determinations made from the fresh plant material, it is found that the yellow pepper, red pepper, broccoli and Brussels

The status of hemostasis at the puncture site of blood access was investigated after the end of hemodialysis. The distribution of hemostatic time at the

Volume 5, Issue 3 available at www.scitecresearch.com/journals/index.php/jrbem/index 595 On the other hand, regardless of the existence at international and national

The aim of the present study was to test for an association between the CYP1A2 − 1545C &gt; T (rs2470890) polymorphism and side effects in a larger sample of patients during

AP: Anteroposterior; CI: Confidence interval; DSCS: Dual SC Screws; ESL: Early sliding length; FNFs: Femoral neck fractures; FSL: Final sliding length; IF: Internal fixation; OR:

Although the economic trend in most target countries was worse than expected and growth in certain product markets was mostly modest, United Internet AG continued its

It is shown by simulation that the CPOS algorithm can derive a set of optimal parameters of WSVM, and WSVM possess some advantages such as fast convergence speed and

Defendants were subpoenaed to appear before a federal grand jury. Antic- ipating that the subpoenaed witnesses would assert the fifth amendment priv- ilege against