Next: Traffic Analysis
Previous: Toolbars

Topic-Sensitive PageRank

Described at the May 2002 World Wide Web (WWW) proceedings, Topic-Sensitive PageRank (TSPR) was developed by Taher H. Haveliwala at Stanford University, the same establishment that spawned Google. Dr Haveliwala was hired by Google in October 2003. TSPR uses search query information to influence link based scores. Unlike PageRank, an inbound-link only counts if it is deemed to be on-topic. Unlike algorithms such as Hilltop it requires minimal real-time processing and does not require a corpus of expert pages related to the search keyword(s).

Like the original Google PageRank algorithm, TSPR computes ranking scores during the indexing of web pages. However it computes multiple ranking scores with respect to various topics. In the original study the topic areas were taken from a top level category of the Open Directory Project. For example, for computing these could be:

Artificial Intelligence
Desktop Publishing
Hacking
Multimedia
Open Source
Operating Systems
Programming
Virtual Reality

Topics are more general than keywords. At query time these scores are combined relative to the search query to form a composite PageRank. As with PageRank this score is used in conjunction with other factors to produce a final ranking.

For an inbound-link to count under the TSPR algorithm it the linking page's theme must be related to the query.

Topic-Sensitive PageRank

See Also