Search tool aids browsing
Technology Research News
Many research teams are working on the problem of how to make finding information on the Web faster and easier. Researchers from Carnegie Mellon University have devised a scheme that gives existing search engines some extra help.
The software, dubbed ScentTrails, shows a user how strongly the links generated by a Web search correlate with the topics she is searching for. The software grades the links a search engine returns by increasing the font size of links that have more connections to relevant pages.
Like conventional search engines, the software uses content cues on a page to determine how useful that page is in relation to a user's query. But the scheme takes searching a step further by showing the user how many other relevant pages a given link is connected to. "A very strongly highlighted hyperlink indicates that many nearby pages match the query closely," said Christopher Olston, an assistant professor of computer science at Carnegie Mellon University.
The software guides the user toward information that matches his search criteria in a way that allows for continuous browsing, said Olston. For example, if a user is looking for a photocopier that copies at least 75 pages per minute, pages that contain links for fast printers will appear larger. This serves to guide the user to select links that are likely to take him closer to his goal rather than links that go cold; this cuts down on the number of times he must interrupt the browsing process to go back and check another link on a search results page.
The idea is to provide "search-driven guidance as people browse the Web, which they are free to follow or ignore as they see fit," said Olston. "The goal is to provide a happy medium between unassisted browsing, which can be tedious, and standard keyword search, in which it is difficult to remain oriented."
The researchers' prototype works within a single site, but could eventually be used in Web-wide searches, and could be used in tandem with existing search engines, said Olston.
The researchers took their prototype through a limited test run using 12 subjects. The tests showed that users were able to consider browsing cues and search cues simultaneously, and the statistics the researchers collected from the study showed that the software allowed the users to locate information more quickly than by either searching or browsing alone, said Olston.
The subjects were given eight tasks. The first three were relatively straightforward: finding a photocopier with recyclable toner, photo support or glossy print capability. The other five involved finding combinations of functions: a digital, black-and-white copier capable of rotating copies, a copier with remote diagnostic technology that prints at least 80 copies per minute, a 400-dots-per-inch copier with a counterfeit deterrence system, a black-and-white copier that scans, faxes, prints and collates, and a 5-to-20-copies-per-minute machine that has photo support.
The users were trained on the software for five minutes, then asked to carry out each task once. Ten of the subjects said they preferred the ScentTrails interface to both browsing and searching for these types of tasks; the two other subjects said they could not tell which was easier.
The main challenge in building the software was making it fast enough. "A naive implementation would take several minutes to determine highlighting gradations for a single Web page, even for a relatively small Web site," said Olston. To make it usable, however, the researchers needed to make the process happen in less than a second, he said.
To do so, the researchers made sure as much information as possible was computed in advance. The software computes the connectivity scores for all pairs of pages on the site ahead of time, said Olston. Match scores, which represent the degree of relevance of pages to queries, are largely computed ahead of time."When a user issues a search query or views a new Web page, highlighting can be performed very quickly by looking up the appropriate set of connectivity and match scores and combining them." Olston said.
The researchers are currently working on three major issues that need to be solved before the software can be used practically across multiple sites, said Olston.
"First, we need to find good methods for highlighting hyperlinks that can be applied across a diverse variety of Web pages, and do not unduly impact content readability," he said. "If link highlighting is too annoying or obtrusive, people will turn it off."
The researchers are also working on improving the software algorithms to make it possible to apply them to very large Web collections, said Olston. "Ultimately I hope ScentTrails or some variant can be used on the Web as a whole, not just on a site-by-site basis."
Third, the researchers are working on finding good starting points for users' queries. "In other words, if someone [has] a partially-formed search query, where should the computer suggest that they begin browsing to find the answer," he said. This would make it easier to apply the method to the entire Web, said Olston.
The method could be used to search within a Web site in one to
two years, and for Web-wide searching in two to five years, according
Olston's research colleague was Ed H. Chi. The work appeared in
the September, 2003 issue of the Association of Computing Machinery (ACM)
Transactions on Human-Computer Interaction. The research was funded
by Palo Alto Research Center (PARC).
Timeline: 1-2 years, 2-5 years
TRN Categories: Internet; Databases and Information Retrieval
Story Type: News
Related Elements: Technical paper, "ScentTrails: Integrated Browsing and Searching on the Web," Association of Computing Machinery (ACM) Transactions on Human-Computer Interaction, September, 2003
March 10/17, 2004
Red wine mends solar cells
Search tool aids browsing
Tiny pumps drive
X-shape pulses hold together
makes tiny scope
Atom spouts photons
Charges make micro
Research News Roundup
Research Watch blog
View from the High Ground Q&A
How It Works
News | Blog
Buy an ad link