August 7, 2006 12:49 PM PDT

Microsoft shows off improved search

A correction was made to this story. Read below for details.
Microsoft scientists have some new thoughts when it comes to search ranking and retrieval accuracy.

New techniques for analyzing search relevance--the way users browse and click through specific content--can improve the algorithms used to rank results, according to two papers submitted by a group of Microsoft researchers.

"Most search engines today use a somewhat two-dimensional approach, matching user queries with the content and link structure of Web pages to return a list of results. We're looking at how to add a third dimension--the users themselves--to improve the search experience," said Eugene Agichtein, a researcher in the Mining, Search and Navigation Group within Microsoft Research.

Thirteen Microsoft papers--10 of them collaborations with university researchers--will be presented at the 29th Annual International Association for Computing Machinery's Special Interest Group on Information Retrieval (ACM SIGIR 2006), which started on Sunday and continues through Aug. 11 in Seattle.

One of the papers, "Learning User Interaction Models for Predicting Web Search Results Preferences," illustrates new techniques for following users beyond the first click-through from a search results page.

The researchers applied factors such as the amount of time spent "dwelling" on a given page, the page with the most "dwell time" for a user given the user's average dwell time per page, and dwell time for pages that rank high for sharing search words with their domain. The researchers looked at the fraction of words shared between search terms, and the domain names, page titles or summaries clicked on by the searcher.

"Using the 'wisdom of crowds' can give us an accurate interpretation of user interactions, even in the inherently noisy Web search setting. Our techniques allow us to automatically predict relevance preferences for Web search results with accuracy greater than the previously published methods," the Microsoft research group said in its paper.

A second paper presented by the same group of researchers discusses how to apply that user information to improve search algorithms. The researchers used an evaluation of 3,000 searches and 12 million user interactions to show that they could improve Web-search-ranking algorithms by 31 percent with their methods.

Other ACM SIGIR search topics focus on translating large amounts of content into smaller, digestible forms. A paper to be presented by researchers from Microsoft, Stanford University and Columbia University covers a "composition context sensitive" system for summarizing multiple documents. A team from the University of Cyprus and the University of Michigan is slated to discuss presenting text summarizations of news stories in formats better suited for mobile devices.

Worldwide, Microsoft's Web browser, Internet Explorer, has a usage rate of 83.05 percent, according to the Web analytics company OneStat.com. Mozilla's Firefox comes in second at 12.93 percent. Search giant Google, the leading search engine in the U.S., has paired with Mozilla on its Firefox Internet browser. Microsoft, whose search engine is MSN Search, recently teamed up with search competitors such as Google and Yahoo in going after click fraud.

 

Correction: This story incorrectly stated the number of collaborative papers Microsoft is presenting at ACM SIGIR 2006. Only 10 of Microsoft's 13 papers are collaborative efforts with universities.

See more CNET content tagged:
researcher, paper, algorithm, Internet search, search engine

3 comments

Join the conversation!
Add your comment
Should MSN improve search or should the user improve search?
If the user is the focus, then, perhaps the burden of GOOD results should be placed on the user.

Currently, search engines take the user's entered keyword and search it within their archives and return a garbled mess. The SERP is focused on the archive, not the user, due to a lack of information provided by the user.

For instance, if the user enters
'widget', the search query should ask, 'do you want to buy a widget', 'do you want detailed information about a widget', 'do you want to compare widgets', 'do you want to learn how to make a widget', 'do you want to buy/read a book about widgets' ... etc.

Of course, this would require the user to be smarter than the widget and we all know that THAT isn't the case. Hence, the burden of good search is placed on the likes of Bill Gates and Company.

Perhaps that is why Mr. Gates is stepping out and using his well-earned monies to espouse education?

Now that's a novel way to enhance our search experience!

Kind Regards,
Al Toman
Posted by vladimirallen (1 comment )
Reply Link Flag
Seeing is believing...
With a search product that sucks as bad as Windows Search and MSN, I doubt anyone will take them seriously until they can prove it.

I can't get Windows search to find a file even if it is the only file in the folder I am search AND I added the search term in the filename and in the file. That is BAD!
Posted by umbrae (1073 comments )
Reply Link Flag
Search
I can see some problems with letting the time someone dwells on a page influence a search result. One it would tend to return the same type of information instead of new information. Suppose a page has more words on it and takes longer to read? How do you know if the person is reading the articles or the ads? How do you know if they are reading or watching TV or talking to someone in the room? What if they get an instant message? How about pictures? Are they looking at pictures or reading? How do you determine the content of the pictures? There are usually more than one section of information which one are they looking at? Unless the software can read minds or watch the user I don't see where this would have a desired result.
Posted by stevenmcs (47 comments )
Reply Link Flag
 

Join the conversation

Add your comment

The posting of advertisements, profanity, or personal attacks is prohibited. Click here to review our Terms of Use.

What's Hot

Discussions

Shared

RSS Feeds

Add headlines from CNET News to your homepage or feedreader.