TrustRank as an inverse, or back-propagation, of PageRank
I’ve read a lot on TrustRank, but I’ve yet to see an analysis of it that I felt confident in because it seems each explanation is more complicated than it needs to be. If Google asked me to design a TrustRank algorithm, here’s how I would do it:
First, PageRank is not as much an invention as it is a discovery, in as much as Einstein’s famous E=mc2 is a discovery of the relationship between mass and energy, and not merely an invention of something that never existed before. PageRank is a discovery if the inherent value of documents based on those documents’ citation history. The more pages that cite your work, the more valuable your page is considered to be, and the more pages that cite the pages that link to your page, the more value those pages give yours. It’s a simple iterative formula that prior to computers would have been impossible to calculate for anything but the smallest web-graph. Since its inception it has distinguished itself as the single most powerful method of getting order from the Internet.
Second, PageRank already has a dampening factor built into it. That’s the (1-d) in the formula, where “d” is generally considered to be 0.15, thus allowing pages to pass on 85% of the value they have. This value can be varied, as if turning a dial, and the returned pages in the SERPs compared to see what value supplies the best results. This dampening factor is thought to be applied universally — equally to all pages across the board, but what if, instead, the dampening factor could be calculated for each individual page, much as PageRank itself is calculated. How would one go about doing this?
PageRank works in one direction, it is the propagation of document value towards a target document, based on the value of the documents linking to it, and the value of the documents linking to those documents that link to the documents that link to the target document, with each document’s passable value divided amount the outgoing links on the page.
Would it be possible to simply reverse the propagation of the algorithm using a variant of the original algorithm such that the value vector pointed back to the linking site instead of the linked-to site? In this way, if you linked to sites of low value and the sites you linked to linked to sites of low value, then it would bring down your TrustRank score.
The beginning dampening factor for the TrustRank algorithm would be a normalized PageRank value. Then when the TrustRank value is calculated, this is then applied to each site in an individual basis as the initial dampening factor value for calculating PageRank.
In this way, pages with high TrustRank are allowed to pass on most or even all the PageRank they have, by having low dampening factors, and sites with low TrustRank would have high dampening factors and thus pass on little of their PageRank value.
Surely, the obviousness of this methodology is not something that would escape the Google engineers, and because of the name being such close kin to the original PageRank nomenclature, surely these methodologies must be closely related. Surely they must not be ignorant of the Linnaean System of biological classification, of genus and families, and as such, TrustRank and PageRank must be more closely related than anyone is saying.
October 13th, 2006 at 19:59
I’m not as sophisticated as those who are able to articluate the in’s and out’s of page rank, trust rank, and so on. I do believe however that there’s really no new ideas being presented out there. All of the “new” ideas are simply iterations of the same core idea – a sites “Rank” is base on its popularity.
I’ve written numerous posts on my marketing blog about increasing page rank and the like. The best comment I ever received was from an individual who said that the beauty of page rank is that you can “change your vote”. I guess what he meant in an overly simplistic way is that PageRank takes so many factors into consideration by focusing on popularity that its a concept that goes way beyond a simple number. ‘nuf said!