3f81f071b8d0024

http://nhiro.org.s3.amazonaws.com/3/5/35bbbc250f5e02c739b3f15f9808ed78.jpg https://gyazo.com/35bbbc250f5e02c739b3f15f9808ed78

(OCR text)

(追記)

Model

Error rate

Error rate

(Positive/

Negative)

18.2 %

(Fine-

grained)

59.0%

Naïve Bayes

(Socher et al., 2013b)

SVMS (Socher et al., 2013b)

Bigram Naïve Bayes

(Socher et al., 2013b),

Word Vector Averaging

(Socher et al., 2013b)

Recursive Neural Network

(Socher et al., 2013b)

Model

Error rate

20.6%

16.9%

BoW (bnc) (Maas et al., 2011)

BoW (bAt'c) (Maas et al., 2011)

LDA (Maas et al., 2011)

Full+BoW (Maas et al., 2011)

Full+Unlabeled+BoW (Maas et al., 2011)

WRRBM (Dahl et al., 2012)

WRRBM BoW (bnc) (Dahl et al., 2012)

MNB-uni (Wang & Manning, 2012)

MNB-bi (Wang & Manning, 2012)

SVM-uni (Wang & Manning, 2012)

SVM-bi (Wang & Manning, 2012)

NBSVM-uni (Wang & Manning, 2012)

NBSVM-bi (Wang &Manning, 2012)

Paragraph Vector

12.20%

11.77%

32.58%

19.9%

11.67%

11.11%

17.6%

12.58%

10.77%

Matrix Vector-RNN

17.1%

16.45%

(Socher et al., 2013b)

Recursive Neural Tensor Network

(Socher et al., 2013b)

Paragraph Vector

NaiveBayes+SVM, bigram*

13.41%

14.6%

13.05%

10.84%

12.2%

11.71%

8.78%

7.42%

むしろBigram NaiveBayes系の優秀さが際立つ?

14F6H6B A