Search is about more than search-engine algorithms. It also must take into account user click-through and browsing behaviors, if search-result accuracy is to be improved.
That’s according to Microsoft researchers, who are set to present details regarding some of the analysis techniques that Microsoft Research is developing to help improve search-result quality. The Microsoft researchers are set to present three papers on this topic at this week’s ACM Special Interest Group on Information Retrieval (SIGIR) conference in Seattle.
By accurately modeling and interpreting user’s search behavior, search engines also can better detect click-spam and deliver more accurate and personalized results, according to the Microsoft researchers.
“It’s not just about a little rectangle on a screen,” said Susan Dumais, a principal researcher with the Microsoft Research Adaptive Systems & Interaction Group. Dumais, a nine-year Microsoft veteran, is one of the authors of the three aforementioned SIGIR papers. “The algorithm is one part of search, but search is also about the user. And it’s not just about Web search – it’s also about searching inside a company and on a PC.”
Microsoft Research’s search work is not purely theoretical. Microsoft announced in January 2006 the formation of LiveLabs, a joint partnership between Microsoft Research and MSN designed to bring search research to commercialization more quickly. (In its latest reorganization, Microsoft announced on August 2 that it was moving LiveLabs to report directly to Chief Software Architect Ray Ozzie.)
Microsoft Research’s search researchers also have been working with the Windows Live Search and Windows Desktop Search teams, Dumais said. Researchers have been investigating for the past few years how a capability like implicit query – or the embedding of relevant search results directly into applications so that users don’t need to pull up a browser and launch a separate query — could fit into Microsoft products, such as Outlook.
Some of Microsoft Research’s search work also is being integrated into Windows Vista and the Windows Live Search Toolbar, Dumais said.
“We’ve been focusing on improving people’s ability to find things they’ve seen before,” Dumais said. “That code has evolved into Vista’s search and browse capability, as well as into the (MSN) Search Toolbar.”
At SIGIR this week, Dumais and other Microsoft researchers will present 13 papers. Dumais was involved in three of them: “Improving Web Search Ranking by Incorporating User Behavior Information”; “Learning User Interaction Models for Predicting Web Search Result Preferences”; and “Improving Personalized Web Search Using Result Diversification.” The focus of all three papers, in short, is how users interact with search results.
Having access to huge volumes of user data is key to improving search results, Dumais said. However, in analyzing this data, “click-through is really important, but the data is really noisy,” Dumais explained.
Microsoft researchers looked at three classes of search-result data, she said: How queries overlap with results (presentation behavior); how many results are found and how often do users click on them (click behavior); and what do users do once they click on a result (browsing behavior). All three categories of data are useful in improving results of users’ search queries, Dumais said.
In the result-diversification paper, Microsoft researchers will highlight learnings in personalized search. Researchers will describe how past behavior, as exemplified by desktop search and history results, can be used to re-rank search results to make them more relevant.
Going forward, “I think we’ll be seeing innovations in users better articulating their searches, with searches (targeting) more specifically certain metadata,” Dumais said. And “if search still looks like a little box in five years, I should be fired,” Dumais said.