Chris Burges, and co-authors, will receive the Test of Time Award for the 2005 paperthat shows how the system works. Burges came to Microsoft Research in 2000, after deciding he wanted to work on machine learning projects that would have an impact on society.
The system, called RankNet, was a breakthrough at the time because it managed to rank results much faster and more accurately than the previous technology. The findings in the paper suggest that it made as much progress training the search engine system to rank results in one day, using one PC, as a previous system had done in several days using a cluster of computers.
The team of researchers tested data consisting of a set of search queries, and for each query, a set of returned documents. In the training phase, some query and document pairs were labeled for relevance such as "excellent match," "good match," and so on. Documents returned in a specific query were ranked against each other.
RankNet relies on neural networks, a computer system modeled to resemble the human brain and nervous system. They are trained to perform tasks based on data labeled by humans.
Microsoft said that other search engines and systems have made use of neural networks, from image capturing to real-time translation. It defines a new era of search engine intelligence that marketers find in platforms like Cortana. Machines are being trained to understand how a human understands images.
Based on visuals in the image, the image-capturing technology within the computer can accurately recognize that it should focus on the dog in the photo, rather than the car, based on a variety of factors such as the words in the article accompanying the image or the metadata in the photo.
Google's PageRank, an algorithm to rank Web sites in Google search results, was named after co-founder Larry Page. The company says users can turn off and on PageRank to see the importance of the ranking on a page.