Statistical Learning in Web Search

Search is becoming the major means for people to access the information on the Internet. According to a survey, 55% of web users use search engines every day. Web search engines are built with technologies mainly from two areas, namely, large-scale distributed computing and statistical learning. Statistical learning is useful because there are many uncertainties in crawling, indexing, ranking, and serving of Web search and the solutions have to be data-driven. In this talk, I will explain how statistical learning technologies are being used in web search. I will also introduce some of the statistical learning technologies for web search, which we have developed recently at MSRA. They include BrowseRrank, ranking refinement, query dependent ranking, and query refinement.