Grouping hyperlinks for improved voice/mobile accessibility
Alex Penev
UNSW / NICTA (Australia)
Accessibility
Heavily presentation-oriented web blind users “read” sequentially
layout gets in the way
Accessibility
Heavily presentation-oriented web blind users “read” sequentially
layout gets in the way
mobile users have small-screen devices For some users, “less is more”
Hyperlinks
Often visit a site with intention of following a link
Hyperlinks
Often visit a site with intention of following a link
Idea: cluster links partition a page’s links into topical
groups helps find a link (cf. IVR) way to organize a set of links
Example: Centrelink.gov.au
Government agency, human services Population obtains info, forms, etc. …but do not visit regularly
information need
Centrelink
111 links on the page Some problems for blind users:
pop-up menus flatten it?
Centrelink
111 links on the page Some problems for blind users:
pop-up menus flatten it?
Some problems for mobiles: hard to fit/navigate a 3-tier pop-up menu
Link clustering
Group related links together hopefully in topical groups
Link clustering
Group related links together hopefully in topical groups
How? label links with descriptions partition descriptions assign summary label to each cluster
Link clustering
Navigation process: invoke clustering step presented with clusters
advertise a label
Link clustering
Navigation process: invoke clustering step presented with clusters
advertise a label
select best-sounding cluster layout is lost
Faster than serial reading
Forming descriptions
A set of keywords for each link anchor text, nested tags, URL, content
Forming descriptions
A set of keywords for each link anchor text, nested tags, URL, content
further possibilities: anchor text window, folksonomy tags, <meta>, ontology
Example australia.gov.au
australia, gov, government, website, travel, taxes, business, people, immigration, etc.
Forming descriptions
Descriptions can be short and have little overlap
Solution?
Forming descriptions
Descriptions can be short and have little overlap
Solution? take into account layout by spreading a
link’s influence across DOM borrow descriptions from neighbor links decay weight with distance
Clustering step
Propagation enriches descriptions relates neighboring links less sparse feature vectors
good for clustering!
Use a partitioning algorithm e.g. k-means
Possible uses
Blind users navigate to link via clustering
skip layout, skip uninteresting topics general surfing?
Possible uses
Blind users navigate to link via clustering
skip layout, skip uninteresting topics general surfing?
Mobile users portal pages
Sample results - Centrelink
12 clusters agency, site, main sections, welfare
allowances, study & training, health, carers, family, children, maternity & job search/retirement, farming, online security
Full table in paper
Conclusion
Simple, well-known ideas applied to a different problem VSM (tf.idf), clustering
May help voice/mobile browsing Future work
what is a good description? what is a good label?