A G: C

unknownlippsΤεχνίτη Νοημοσύνη και Ρομποτική

16 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

96 εμφανίσεις

A
CCESS

G
RADE
:
C
ROWDSOURCING

THE

W
EB

A
CCESSIBILITY

P
ROBLEM


WWW
.A
CCESS
G
RADE
.
COM

#
A
CCESS
G
RADE

Presenter:
Sina

Bahram

(@
SinaBahram
)


Collaborators:
Srinath

Ravindran
,
Robert St.
Amant
, and Denis
Bahler


North Carolina State University

Department of Computer
Science

Raleigh, North Carolina (USA)

W
HAT

A
RE

W
E

G
OING

TO

T
ALK

A
BOUT
?


Problem
: Assessing Web Accessibility


Proposal:

Automated Approach?


Explanation:

Techniques

Used So Far


Results:

Preliminary Data


Contribution:

Dataset, Features, and Method


Future Work:

Automatic Transcoding


Conclusion:
Final Thoughts


P
ROBLEM
: A
SSESSING

W
EB

A
CCESSIBILITY


Assessing accessibility of websites is time
consuming


It mostly requires humans to be done right


When problems
are found,

humans are needed to
fix them


Both assessment and resolution of problems can
be very costly


Accessibility here refers to compliance with
standards

P
ROPOSAL
: A
UTOMATED

A
PPROACH
?


Can we use machine learning techniques to
automate:


Assessment
of accessibility
?


Explanation
of accessibility problems when
they
exist?


Solutions
to said accessibility problems
?


Tailoring specific results to an individual user?


Generalizing from these results to help all users?


E
XPLANATION
: T
ECHNIQUES

U
SED

S
O

F
AR


Steps
:


1
. Gather samples of accessible and inaccessible
websites


2
. Break down each page into structural
components


3
. Input data into decision tree, Bayesian network,
and support vector machine
classifiers


4
. Use classifiers to predict accessibility of a webpage

R
ESULTS
: P
RELIMINARY

D
ATA


Binary Classifier


Accessible or Inaccessible


Success rate: 85%



Multi
-
Class Classifier


Sliding scale of accessibility


Success rate: 60
-
65%


A majority of misclassifications are only 1 off

C
ONTRIBUTION


A dataset of webpages marked as accessible or
not


A dataset of web pages marked with a score of 1
-
5 in terms of accessibility (data from the crowd)


An automated way of discovering features
that
are predictive of accessibility


A new method for assessing web based
accessibility

F
UTURE

W
ORK


Structural
-
based
commentary of website
accessibility


Automatic transcoding of inaccessible
web pages
in real time


Tailoring the above to the individual user


Extracting features common to most/all users

C
ONCLUSION


Basing assessment of web accessibility on
structural components leads
to:


A
more systemic/generic
approach


The
ability to classify large quantities of
webpages


The ability to offer
specific suggestions to content
authors


The ability to
p
rovide real
-
time
transcoding of
inaccessible content in the future

Q
UESTIONS
?