Microsoft Researchers Inventing New Techniques to Improve Search Engine Accuracy and Relevance

SEATTLE — Aug. 7, 2006 — Scientists at Microsoft Corp. are developing new techniques for analyzing search click-through patterns and browsing behaviors to make search results more relevant. Microsoft researchers Eugene Agichtein, Eric Brill, Susan Dumais and Robert Ragno report that accurate modeling and interpretation of user interactions with a search engine can significantly improve search-result ranking, the detection of “click-spam,” Web search personalization and, ultimately, the overall Web search experience. While user interactions with the Web search engines are plentiful, new robust techniques are required to understand the relationship between user interactions and result quality.

The paper detailing this research, “Learning User Interaction Models for Predicting Web Search Result Preferences,” represents one of 13 papers to be presented by Microsoft Research at the 29th annual ACM SIGIR (Association for Computing Machinery’s Special Interest Group on Information Retrieval) conference on search and information retrieval in Seattle this week. Microsoft Research contributed the largest number of papers to this year’s conference, presenting 17.5 percent of the 74 papers accepted out of a record 399 submissions. SIGIR is a top international forum for the presentation of new research results and the demonstration of new systems and techniques in the broad field of information retrieval.

Another Microsoft Research paper by the same authors, “Improving Web Search Ranking by Incorporating User Behavior,” discusses utilizing the discovered patterns of user behavior to improve algorithms that rank search engine results. This work is significant because it demonstrates how to add a new rich set of user behavior features into ranking for large relevance gains.

“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,” said Eugene Agichtein, a researcher in the Text Mining, Search and Navigation Group within Microsoft Research. “We’re looking at how to add a third dimension — the users themselves — to improve the search experience. By examining click-through and browsing patterns across a large number of users, we are able to learn a great deal about how people interact with search technologies and can thereby improve our accuracy dramatically.”

The work Microsoft Research is presenting at this year’s conference spans well beyond user behavior research, however, covering areas such as feedback relevance, cross-language retrieval, query analysis and classification, summarization, personalization, graph structure analysis, and the development of new machine learning algorithms for search.

Continuing Microsoft’s long history of collaboration with the academic community to encourage research and innovation, the majority of the work Microsoft is presenting at SIGIR 2006 was done in cooperation with researchers at academic institutions around the world. Ten of the 13 papers the company is presenting were co-written by researchers at major universities. This follows the company’s June 2006 announcement that it was making available 12 research grants and rich query logs and usage data to help support academic research and improve Internet search, data mining, discovery and analysis technologies.

Microsoft also has a long history of involvement in the search community and SIGIR. Search and information access have been central themes at Microsoft, reaching as far back as 1993. Microsoft Research has been active in the SIGIR community since 1997.

“Microsoft’s involvement in the SIGIR and the search communities is extensive,” said Efthimis Efthimiadis, chairman of this year’s SIGIR conference and associate professor at the University of Washington Information School. “In addition to openly publishing research papers in increasing numbers year after year, Microsoft researchers can be found serving on executive boards, chairing committees, reviewing papers and presenting keynote addresses at conferences. I believe the work Microsoft Research is doing is advancing the field of search, and it will improve the online experience for all users.”

Links to the papers Microsoft Research contributed to this year’s SIGIR conference, as well as to the Microsoft papers presented at the 2004 and 2005 conferences, are available at

About Microsoft Research

Founded in 1991, Microsoft Research is dedicated to conducting both basic and applied research in computer science and software engineering. Its goals are to enhance the user experience on computing devices, reduce the cost of writing and maintaining software, and invent novel computing technologies. Researchers focus on more than 55 areas of computing and collaborate with leading academic, government and industry researchers to advance the state of the art in such areas as graphics, speech recognition, user-interface research, natural language processing, programming tools and methodologies, operating systems and networking, and the mathematical sciences. Microsoft Research employs more than 700 people in five labs located in Redmond, Wash.; Silicon Valley, Calif.; Cambridge, England; Beijing, China; and Bangalore, India. Microsoft Research collaborates openly with colleges and universities worldwide to enhance the teaching and learning experience, inspire technological innovation, and broadly advance the field of computer science. More information can be found at

About Microsoft

Founded in 1975, Microsoft (Nasdaq “MSFT”) is the worldwide leader in software, services and solutions that help people and businesses realize their full potential.

Note to editors: If you are interested in viewing additional information on Microsoft, please visit the Microsoft® Web page at on Microsoft’s corporate information pages. Web links, telephone numbers and titles were correct at time of publication, but may since have changed. For additional assistance, journalists and analysts may contact Microsoft’s Rapid Response Team or other appropriate contacts listed at

Related Posts