Examining Variations of Prominent Features in Genre Classification



distinguish most pairs which include Periodicals which
conforms to instinct, since periodicals deal with a wide
range of topics, and we do not expect Rainbow features
which emphasise topical distinction to fare well. In
particular, visually elaborate periodicals and structurally
formal theses are unsurprisingly best distinguished by
image features.

The same consideration of confusion levels
indicates that topical features do little to distinguish genre
classes likely to have similar topic areas such as Academic
Monograph and Thesis. And it supports the expectation
that distinctions between Academic Monographs and
Book of Fiction, and Book of Fiction and Thesis would be
highly topical.

Table 9. The difference between maximum and
minimum belief and error impact confusion.

Metric

Genre pair

CB

CE

AM

BF

0.25

0.01

AM

BR

0.5

0.06

AM

M

0.2

0.51

AM

P

0.2

0.08

AM

T

0.08

0.42

BF

BR

0.39

0.22

BF

M

0.08

0.54

BF

P

0.13

0.66

BF

T

0.32

0.73

BR

M

0.14

0.18

BR

P

0.37

0.09

BR

T

0.07

0.28

M

P

0.68

0.24

M

T

0.08

0.64

P

T

0.07

0.39

The difference between the maximum and
minimum confusion level seen across features between
pairs ranges from 0.01 to 0.73 (on the basis of
CE), and,
from 0.07 to 0.72 (on the basis of
CB) [see Table 9].
Although the features indicated in Table 8 are the features
exhibiting the lowest confusion levels, in some cases, the
difference is very slight (e.g. Academic Monographs and
Book of Fiction). In interpreting the information in Table
8, it seems reasonable to take the differences noted in
Table 9 into consideration. For example, the pair of
classes Book of Fiction and Periodicals has been
examined to be best distinguished by style and image (see
Table 8), but the figures in Table 9 seem to suggest that
the weight is more prominently on image. Likewise,
Academic Monographs and Minutes seem best
distinguished by style and Rainbow with a higher weight
placed on Rainbow.

Further feature strengths across pairwise
classification are observable in Table 10, where the
contents of Tables 8 and 9 have been merged for a
convenient overview.

Table 10. Feature strengths in pairwise classification.

Genre pair

Metric

CB

CE

AM

BF

0.25

style, Rainbow

0.01

Rainbow

AM

BR

0.5
style

0.06
style

AM

M

0.2
style

0.51

Rainbow

AM

P

0.2
style

0.08
style

AM

T

0.08
style

0.42
image

BF

BR

0.39
style, Rainbow

0.22
style

BF

M

0.08
style

0.54
style

BF

P

0.13
style

0.66
image

BF

T

0.32
style, Rainbow

0.73
Rainbow

BR

M

0.14
style, Rainbow

0.18
Rainbow

BR

P

0.37
style

0.09
style

BR

T

0.07

style, Rainbow

0.28
style

M

P

0.68
style

0.24
image

M

T

0.08
style

0.64
style

P

T

0.07
image

0.39
image



More intriguing information

1. Performance - Complexity Comparison of Receivers for a LTE MIMO–OFDM System
2. The name is absent
3. Should informal sector be subsidised?
4. BARRIERS TO EFFICIENCY AND THE PRIVATIZATION OF TOWNSHIP-VILLAGE ENTERPRISES
5. The name is absent
6. FASTER TRAINING IN NONLINEAR ICA USING MISEP
7. The name is absent
8. Public infrastructure capital, scale economies and returns to variety
9. Political Rents, Promotion Incentives, and Support for a Non-Democratic Regime
10. Measuring Semantic Similarity by Latent Relational Analysis