Microsoft aims to boost search engine accuracy

Microsoft researchers yesterday unveiled newly developed techniques designed to improve analysis of search click-through patterns and user browsing behaviours.

Redmond scientists Eugene Agichtein, Eric Brill, Susan Dumais and Robert Ragno reported that accurate modelling and interpretation of user interactions with a search engine can significantly improve search-result ranking, the detection of 'click-spam', web search personalisation and, ultimately, the overall internet search process.

While data investigating 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 researchers argued.

The research is covered in a paper, Learning User Interaction Models for Predicting Web Search Result Preferences, which is due to be presented 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.

Another Microsoft Research paper by the same authors, Improving Web Search Ranking by Incorporating User Behaviour, discusses using the discovered patterns of user behaviour to improve algorithms that rank search engine results. This work demonstrates how to add a new set of user behaviour 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."

Links to the papers Microsoft Research contributed to this year's SIGIR conference are available here.