Complementing Standards by Demonstrating Commitment and Progress—W4A 2015
W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility
-
Upload
rui-lopes -
Category
Technology
-
view
624 -
download
2
description
Transcript of W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility
![Page 1: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/1.jpg)
Web Not For AllA Large Scale Study of Web Accessibility
Rui Lopes1, Daniel Gomes2, Luís Carriço1
1 LaSIGE, University of Lisbon2 FCCN
![Page 2: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/2.jpg)
Context
• The Web is the biggest information source for Mankind. Decentralised architecture made it blossom.
• Humans (and computers!) contribute to information production and consumption, leading to ~45B Web pages.
![Page 3: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/3.jpg)
Context
• Growth of users contributing and interacting with the Web leads to significant diversity of users, including people with disabilities.
• The openness and decentralisation of the Web leads to an uncontrolled quality check of Websites’ usability (and accessibility).
![Page 4: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/4.jpg)
What is the state of accessibility on the Web?
![Page 5: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/5.jpg)
• It is known that Web accessibility adequacy is often far worse than desired.
• Studies tend to focus on a restricted (small) set of Web sites.
• Do macroscopic properties of Web accessibility emerge from analysing at a large scale?
![Page 6: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/6.jpg)
Experiment
background
• The Portuguese Web Archive initiative periodically crawls contents from the Portuguese Web (.pt and others) for future preservation.
• Services are built on top of crawled collections: search (end users) & analysis framework (researchers).
![Page 7: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/7.jpg)
Methodologydata acquisition - obtaining the document collection
• Collect a sufficiently large portion of the Web, yet representative (e.g., national Webs)
• Spider traps handled gracefully
• Boostraped with 200,000 Website addresses from the .pt TLD
• Collected March/May 2008
![Page 8: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/8.jpg)
Methodologydata acquisition - evaluation process
• Implementation of 39 WCAG 1.0 checkpoints yield pass, fail, warn. (collection previous to WCAG 2.0 TR)
• Overcome computational effort with Hadoop cluster, streams, caching, etc.
![Page 9: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/9.jpg)
Methodologydata analysis
• Failure rate, 3 criteria:
![Page 10: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/10.jpg)
Resultsgeneral
• 28M Web pages were evaluated. (58%)
• 21GB evaluation data collected for analysis.
• 40B HTML elements evaluated. (~1500/page)
• 1.5B elements passed. (56/page, 3.89%)
• 2.9B elements failed. (103/page, 7.15%)
• 36B elements warned. (1291/page, 89%)
![Page 11: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/11.jpg)
Resultsrates versus page count distribution
conservative optimistic strict
![Page 12: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/12.jpg)
Resultsrates versus page complexity (# HTML elements)
conservative optimistic strict
![Page 13: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/13.jpg)
Discussionon the results
• Large scale confirms predictions of small scale studies - the Web is still not for all.
• Smaller Web pages tend to have greater accessibility quality.
• Nature of warnings is more striking than expected, completely different interpretations.
• Automated evaluation is just the beginning.
![Page 14: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/14.jpg)
Discussionon the limitations of the experiment
• HTML structure vs. content rhetorics. (CSS & Javascript can change it all)
• Collecting the Web is hard. (deep Web - AJAX & forms -, infinite generation, robots.txt, etc.)
• Scaling evaluation & analysis processes is hard. (evaluation streamability, resource inter-dependencies, billion node graphs, etc.)
![Page 15: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/15.jpg)
Conclusions• Large scale accessibility evaluation of the
Portuguese Web.
• Re-confirmed studies at the large.
• Educating developers & designers about warnings is crucial for accessibility success!
• Automated evaluation is just the start. Always need for expert & users evaluations.
![Page 16: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/16.jpg)
• Linking properties (ranking vs. accessibility)
• Evolution of accessibility compliance in time (different document collections)
• Cross-cuts: gov, e-com, personalisation, etc.
• Developing countries (Portuguese speaking African countries)
Ongoing Workwe’re still at the tip of the iceberg
![Page 17: W4A 2010 - Web Not For All: A Large Scale Study of Web Accessibility](https://reader035.fdocuments.us/reader035/viewer/2022080211/558c8a55d8b42a63678b4583/html5/thumbnails/17.jpg)
• Making available evaluation datasets (e.g., Linked Data). Ours and yours!
• Larger document collections.
• Transforming warnings into failures with machine learning.
Ongoing Workhelp wanted from community!