Access patterns organize data
By
Kimberly Patch,
Technology Research NewsAfter more than a decade of rapid growth, the World Wide Web has grown to more than 40 million servers. But despite the efforts of many researchers, this abundance of information remains relatively disordered.
Researchers from Old Dominion University are aiming to organize large bodies of information using a method that automatically generates links among digital objects based on the way the human brain organizes information. The human brain contains about 100 billion neurons. "Our methodology mimics how certain parts of the brain learn to make connections between related information," said Aravind Elango, a researcher at Old Dominion University.
The method could eventually allow information repositories like the Internet to self-organize based on the way users access information. Self-organizing digital collections have the potential to cut search time, make searching digital collections more intuitive, and preserve information about the relationships among data, said Elango. "Digital information is being lost now because information is passive. We want to put the information itself in the driver's seat where possible, and give it the capabilities to adapt and maintain its own integrity," he said.
The researchers tested the method using 150 data objects, or buckets that provide information about a music band. Each bucket contains information and also the means to manage it, including methods of interacting with users and maintaining links to other buckets. Buckets are akin to folders of information, but also contain the ability to display information and the intelligence to maintain links to other folders, said Elango.
The researchers set up the buckets using random links, and as users traversed the system links were changed according to Hebb's Law of Learning, which is a rough model of how neurons in the human brain work. Hebb's law postulates that the connection between a pair of neurons becomes stronger when neurons are activated in quick succession. This method is useful for situations in which it is not possible to train the system using a set of correct or incorrect responses in advance, said Elango.
Each bucket keeps track of the two most recently visited buckets and adds new links to those buckets based on users' travels, said Elango. Over time the more relevant links come to occupy higher positions in each bucket.
In this way the system taps the judgment of users traversing the system to rank the relevance of a link to a bucket, said Elango. "As users traverse the system, it learns about user preferences from experience and absorbs the users' knowledge on a subject."
The researchers tested their system on 15 users who collectively took about 1,000 steps around the buckets, according to Elango. "The system showed that after significant user traversal, each bucket contains a set of links [that] are more relevant to the music band," he said. The organization happened even though the people using the system were diverse and not fans of one particular type of music, Elango added.
For example, the bucket containing information about the band The Clash initially contained three random links -- to the Beatles, Glyn Jones and Beck. After the user activity, Smashing Pumpkins was at the top of the Clash bucket's linked list, followed by Beck, Fishbone, Nick Lowe, The Beatles, The Smith's, The Replacements, Glyn Jones, N. W. A., and Squeeze.
The bucket containing information about The Smith's initially contained random links to Elvis Costello, Tool and The Replacements. After the user activity, The Replacements were at the top followed by Elvis Costello, Pretty Things, Fishbone, Nick Lowe, The Beatles, The Clash, Johnny Thunders, Kiss, and Tool.
The demonstration showed that, with sufficient intelligence, digital objects can generate dynamic recommendations for users, said Elango. The network reflects true references rather than accepted notions of how these bands should relate, he added.
Because buckets can contain any information, the system can be used with Web pages, said Elango. The method could be used in Internet recommender systems in a variety of ways, he said. "Examples include making the document collection of a digital archive more easily traversable, or generating recommendations for each individual user."
The method could also be used to study the Internet's characteristics as users would like to see it, said Elango. This is akin to allowing a highway system to adapt to actual traffic patterns rather than laying out roads ahead of time, he said. "The highway system would not only be perfectly adjusted to how people travel, but also be a representation of their travel preferences that we can learn from."
Elango's research colleagues were Johan Bollen and Michael L. Nelson. The research was funded by Old Dominion University.
Timeline: Unknown
Funding: University
TRN Categories: Internet; Databases and Information Retrieval; Computers and Society
Story Type: News
Related Elements: Technical paper, ""Dynamic Linking of Smart Digital Objects Based on User Navigation Patterns" posted on CoRR at arxiv.org/abs/cs.DL/0401029
Advertisements:
June 2/9, 2004
Page One
Recognition keys access
Rules aim to get devices talking
Access patterns organize data
Atom-photon link demoed
Briefs:
Plastic nanowires sense gasses
Process nets cheap microstructures
Cursor speed shows virtual bumps
Sensors track martial arts blows
Nanotube moves molten metal
Buckyballs gain smaller kin
News:
Research News Roundup
Research Watch blog
Features:
View from the High Ground Q&A
How It Works
RSS Feeds:
News
| Blog
| Books 
Ad links:
Buy an ad link
| Advertisements:
|
 |
Ad links: Clear History
Buy an ad link
|
TRN
Newswire and Headline Feeds for Web sites
|
© Copyright Technology Research News, LLC 2000-2006. All rights reserved.
