Scent Trails: Integrating Browsing and Searching on the Web Christopher Olson et al. Blake Adams November 4, 2003.

Scent Trails: Integrating Browsing and Searching on the

WebChristopher Olson et al.

Blake AdamsNovember 4, 2003

Information Location Methods

• Browsing– View pages one at a time,

navigation via hyperlinks

• Drawback– Inefficient, relies on

browsing cues

• Benefit– Useful where keywords are

nonexistent or unavailable

Information Location Methods

• Searching– Enter a search query into a

search engine, choose sites from ranked list of results.

• Drawback– Often returns inappropriate

results, loses important context.

• Benefit– Identifies pages containing

specific information quickly. – Searches are tailored to

users specific needs.

Information Location Methods

• ScentTrails– Automatically high light

hyperlinks to guide users toward search results while still allowing conventional browsing.

– Degree of highlighting based on ‘Information Scent.’

– Considers search and browsing cues together, user can make informed navigational decisions and locate content matching complex information goals.

Scent Trail Goals

• Based on information scent developed in the context of information foraging.– How users interleave

directed structured behavior (searching) with opportunistic and unstructured behavior (browsing).

• The goal on ScentTrails is to assist browsing by adding new supplemental cues tailored to the needs of individual users.

How it Works

• User inputs a list of search terms into an input box, these keywords represent the user’s partial information goal. As browsing/search continues, this box can be modified, and ScentTrails will update dynamically by highlighting existing links that best satisfy the information goal.

• ScentTrails also considers distance to relevant pages, discounting the value of more distant pages accordingly.

Technique

• Identify pages that match the user’s partial information goal.

• Determine distance to desirable results, weigh based on number of clicks necessary to reach goal.’

• Weigh fonts of links based on results.

Algorithm

Evaluation

• Based on a crawl of www.xerox.com• Used 12 Test Subjects who were assigned 8

searching tasks.• Search Options:

– Browsing– Searching– ScentTrails– ShortScent (result pages only crawl neighboring

pages)• Each Test Subject completed each task only

once - a set of 2 tasks with each method. These methods were rotated from user to user.

Tasks Performed

1. Find a copier with recyclable toner2. You have a business at home and you want a copier

with photo support.3. Find a copier with glossy print capability.4. Find a digital, black & white copier that supports

rotation.5. Find a copier with remote diagnostic technology that can

do at least 80 copies per minute.6. Find a 400 dpi copier with a counterfeit deterrent

system.7. Find a black & white machine that supports scan, fax,

print and copy with collation.8. Find a 5 to 20 cpm (copies per minute) copier with photo

support.

Results/Analysis

Average task completion time for each task/interface pair, with standard error.

Results/Analysis

Average task completion time for each of the interfaces combined across tasks, with standard error.

Results/Analysis

Results/Analysis

• Interviews– Interface Preference: 10 out of 12 preferred Scent

interfaces. Dissenting subjects declined to state a preference due to lack of experience

– Several Subjects remarked on how the Scent interface allowed them to narrow down some aspects of the task with keywords, while honing in on the other aspects of browsing.

– Highlighting: Highlighting method was praised for making searches more efficeient, but criticized ambiguity of textual sizing.

Highlighting options explored