Google: Hilltop alogrithm

[Brian Turner]

In my opinion, the Google algorithim known as "Hilltop" is almost certainly implemented post-Florida.

The Hilltop paper is actually available online here:

http://www.cs.toronto.edu/%7Egeorgem/hilltop/

And here's the most interesting part:

We described a new ranking algorithm for broad queries called Hilltop and the implementation of a search engine based on it. Given a broad query Hilltop generates a list of target pages which are likely to be very authoritative pages on the topic of the query. This is by virtue of the fact that they are highly valued by pages on the WWW which address the topic of the query. In computing the usefulness of a target page from the hyperlinks pointing to it, we only consider links originating from pages that seem to be experts. Experts in our definition are directories of links pointing to many non-affiliated sites. This is an indication that these pages were created for the purpose of directing users to resources, and hence we regard their opinion as valuable. Additionally, in computing the level of relevance, we require a match between the query and the text on the expert page which qualifies the hyperlink being considered. This ensures that hyperlinks being considered are on the query topic. For further accuracy, we require that at least 2 non-affiliated experts point to the returned page with relevant qualifying text describing their linkage. The result of the steps described above is to generate a listing of pages that are highly relevant to the user's query and of high quality.


Hilltop most resembles the connectivity techniques, PageRank and Topic Distillation. Unlike PageRank our technique is a dynamic one and considers connectivity in a graph specifically about the query topic. Hence, it can evaluate relevance of content from the point of view of the community of authors interested in the query topic. Unlike Topic Distillation we enumerate and consider all good experts on the subject and correspondingly all good target pages on the subject. In order to find the most relevant experts we use a custom keyword-based approach, focusing only on the text that best captures the domain of expertise (the document title, section headings and hyperlink anchor-text). Then, in following links, we boost the score of those targets whose qualifying text best matches the query. Thus, by combining content and connectivity analysis, we are both more comprehensive and more precise. An important property is that unlike Topic Distillation approaches, we can prove that if a page does not appear in our output it lacks the connectivity support to justify its inclusion. Thus we are less prone to omit good pages on the topic, which is a problem with Topic Distillation systems. Also, since we use an index optimized to finding experts, our implementation uses less data than Topic Distillation and is therefore faster.

The indexing of anchor-text was first suggested in WWW Worm [McBryan 94]. In some Topic Distillation systems such as Clever [Chakrabarti et al 1998] and in the Google search engine [Page et al 98] anchor-text is considered in evaluating a link's relevance. We generalize this to other forms of text that are seen to "qualify" a hyperlink at its source, and include headings and title-text as well. Also, unlike Topic Distillation systems, we evaluate experts on their content match to the user's query, rather than on their linkage to good target pages. This prevents the scores of "niche experts" (i.e., experts that point to new or relative poorly connected pages) from being driven to zero, as is often the case in Topic Distillation algorithms. In a blind evaluation we found that Hilltop delivers a high level of relevance given broad queries, and performs comparably to the best of the commercial search engines tested.

I'm not sure that I quite agree with the final analysis, but it does warn SEO in general that links by themselves should no longer be relied upon in themselves: a particular structuring in ny backlink network should be aimed for.