In November of 2003, Google performed a major algorithm change. The goal of the change was to make it harder to manipulate their search result. It is believed that Google may have a number of algorithm and concept of the following section, including Hilltop, Topic Specific page rank, Trust Rank, Temporal analysis and Latent Semantic Indexing.
Some of the these algorithm may not be part of the current search environment but the idea contained within them are still worth understanding see where search may be headed & what search topic search engineers think are important to improving there are overall relevancy scores.
Hilltop's
Hilltop is an algorithm with reorganizes search result based on an expert rating system.
In the Hilltop white paper they talk about how expect documents can be used to help compute relevance. An expect document is a non-affiliated page which links to many related resources. If page A is related to page B and page B related to page C Then connection between A and C assumed.
Additionally, Hilltop states that it strongly consider page title and page heading in relevance scores. Likely Hilltop also consider the link pointing into the page and site which your links come from.
The benefit of Hilltop over raw page rank (Google) is that it is topic sensitive and is thus generally header to manipulate then buying some random high-power of link. The benefit of Hilltop over topic distillation (Teoma) at that Hilltop is quicker and cheaper to calculate. It is believed that Google might be using Hilltop to help sort the relevancy for some of their search results.
Topic sensitive page rank:
Topic Sensitive Page Rank biases both the query and the relevancy of returned documents based upon the perceived topical conte3xt of the query.
The query context can be determined based on search history, user defined input (such as search personalization - try Googles Labs Search Personalization if you are interested), or related information in the document from which the query came.
Topic Specific page rank for each page can be calculated off line. Using an exceptionally coarse topic set ( for example, the top leave Open Directory Projecst categories) still allows Topic Sensitive page rank to significantly enhance relevancy over using page rank alone, however T.S.P.R can be applied more specifically as well.
Since much of it is calculated off line, Topic Specific page rank can also be rolled into other relevancy algorithms which are calculated in near real time.
Latent Semantic Indexing:
Latent semantic indexing allows machines to understand language by looking at it from purely mathematical view point.
Latent semantic indexing adds an important step to the document indexing process. In addition to recording which keywords a documents contains, the method examines the document contain as a whole, to see which other documents contain some of those same words. LSI considers documents that have many words in common to be semantically close, and ones with few words in common to be semantically distant. This simple method correlates surprisingly well with how a human being, looking at content, might classify a document collection. Although the LSI algorithm doesn’t understand anything about what the words mean, patterns it notices can make it seem astonishingly intelligent.
Latent semantic indexing is a rather expensive process and many SEO experts debate to what extent major search engines may be using the technology. If they are not using it much yet, in time they surely will.
Temporal Analysis:
Search engines can track how long things (sites, pages, and links) have been in existence and how quickly they change. They can track ling a domain has been in existence, how often page copy changes, how page copy changes, how large a site is, how quickly link popularity builds, how any particular link exists, how similar the link text is, how a site changes in rank over time, how related linking sites are, and how natural linkage data looks.
In some cases, it makes sense for resources to acquire a bunch of linkage data in a burst. When news stories about a topic and search volumes on particular term are high it would also make sense that some sites may acquire a large amount of linkage data. Most the time if links build naturally, they build more slowly and evenly.
If links build in hue spikes then search engines may discount- or even apply a penalty- to the domain receiving that linkage data if those links do not build in somewhat regular pattern.
Stale pages may also be lowered in relevancy. A page may be considered fresh if it changes somewhat frequently of if it continues to acquire linkage data as time passes.
Google may also look at how often your site is bookmarked, who your advertisers are, and other various feed back they can get from their toolbar.
Google was awarded a patent on March 31, 2005 covering these types of topics (scout in much more detail).
Trust Rank:
Trust Rank is and algorithm which can be used to bias Page Rank by placing additional authority on human reviewed trusted sites. Trust propagates out from the trusted pages and sites to pages and sites they link to. Trust Rank also can be used to neutralize the effects of some types of low quality link building from un trusted sources, as well as flag high Page Rank low trust websites for human review.