What good is a search engine if the results it provides aren’t relevant to the query being performed? The answer? Not much. The largest search index in the world doesn’t amount to much if you don’t have an algorithm that can successfully provide results related to the question being posed.
Because relevance is such an important characteristic, many who study the industry conduct tests to find out which engine is the most relevant and surveys to discover what people consider relevant when it comes to search results.
Evidently, the commitment to relevance is also of great importance to the developers at MSN Search, who recently began using a new method to determine relevance within their search results called the Neural Net. According to Barry at SERoundtable, the Neural Net technology is based on RankNet, a method of relevance ranking being researched by the MSN team.
We investigate using gradient descent methods for learning ranking functions… and we introduce RankNet, an implementation of these ideas using a neural network to model the underlying ranking function.
This pretty much confuses everyone else as much as it does me. Even Danny Sullivan had a great time deciphering their white paper: “I’d love to give you a one sentence summary of what it does, but so far, that escapes me despite reading the paper several times.”
Danny also feels the RankNet system “recognizes” what is a good result and what is not and ranks them accordingly. He also admits that this line of thinking may come from him misunderstanding their pdf.
While the white paper is quite dense, their use of the Neural Net has already paid dividends, at least in Japanese field tests and other examples provided on the MSN Search blog. Although they do not provide in-depth details about their technology, the post does say the developers at MSN Search are obsessed about relevance and the Neural Net technology should be an indication they are committed to improving the user experience.
To illustrate their technology in action, the blog entry explains how queries about PBS evolution videos have seen relevance improvements in a measurable way. Before they integrated their Neural Net, the first position result was not completely relevant to the search (i.e., PBS did not have the top spot as it should). However, since the integration, the PBS website now holds the first position for the query in question.
Not only has MSN Search made strides to improve their result relevance, they’ve also introduced some new search commands that will assist users on refining their searches. According to their blog entry, the following commands have been added (complete with descriptions explaining how they work):
FileType: one of the most asked for operators, which restricts documents to a particular filetype.
InAnchor:, InURL:, InTitle:, and InBody: available to find keywords in a particular part of the document, or in anchor text pointing to a document.
Augmented the Link: keyword that finds documents that link to a particular page with LinkDomain:
Contains: returns documents that contain hyperlinks to documents with a particular file extension (contains:wma returns documents that contain a link to a WMA file).
With their commitment to search relevance and the ability for users to refine their searches using a number of search commands, it certainly appears as if MSN Search is committed to providing as complete a search experience as possible. Does this new relevance algorithm affect how site owners will approach optimizing their work for MSN Search? When you consider the fact that the MSN Search results will inevitably be altered and rankings will be changed, perhaps it will. Although, exactly which SEO measures can be taken, besides ensuring your site content is relevant to your targeted keywords, is uncertain.
It’s probably wise to give the new ranking algorithm some time to take hold first.