Next: Creation Date Ranking
Previous: Optimizing for Google

Information Retrieval Based on Historical Data

(Google US Patent 20050071741)

On the 31st of March 2005 Google filed a broad patent application. It addresses two common complaints with Google Search, spam and the difficulty new but interesting sites have getting ranked. Both these problems can be traced back heuristic flaws in the original PageRank algorithm.

There are a few things to remember before basing a search engine marketing strategy around this filing. Patents, particularly those covering business processes, are regularly filed by companies to be used as bargaining chips with competitors. We have already seen that Google had to pay Yahoo! who held a patent covering some implementation details of Google Adword program. The processes described may not be used in Google's algorithm today but may be incorporated in the future or may simply have been filed to stop other companies using the techniques.

It is unlikely (although not impossible with the whacky US Patent Office) that all of the "inventions" claimed will be granted patent protection. Some of the ideas are not original and prior art can be found.

As the title suggests, the patent is concerned with how historical data can be used to sort search results for relevance. Google's researchers have pretty much brainstormed all the ways that date information relating to a document can be discovered and have then described some processes that can be used to score documents based on this information either alone or in combination with other mechanisms such as PageRank.

The patent application outlines the methods used by search engines to determine relevance in response to queries, these are on-page and off-page factors such as the PageRank algorithm. It goes on to outline the problems with these systems, including search engine spam and older, higher ranking but stale information. It then lists some of the historical data that may be considered when filtering documents: Document creation and update times; analysis of queries performed on the search engine; link and anchor text criteria; user behaviour (possibly gathered through the toolbar; information connected to the registration and maintenance of the domain name; ranking history (this could cover the speed at which inbound-links are acquired); bookmarking (through the toolbar or browser); unique words, phrases and bigrams used in the anchor text; context related linking.

Information Retrieval Based on Historical Data

(Google US Patent 20050071741)

See Also