CollaborativeRank Says I'm an Expert on XML, Mining and Validation

Martin Dittus · 2005-11-12 · data mining, links, recommendation engines, tools · write a comment

CollaborativeRank is an interesting service that builds on the database. They provide bookmark search, a ranking of popular bookmarks, and they attempt to find connections between the things people store in their account and their area of expertise. It's the last feature that I find the most interesting. While it disguises as a ranking of users, its main promise is that it could help you find experts on arbitrary fields.

During the last couple of weeks I've been watching my rank, and while I wouldn't necessarily agree with its estimation of my expertise it's still interesting to watch.


At the time of writing I'm 120th on CollaborativeRank's list of users.

When I first bookmarked their page on 2005-10-09 my main area of expertise according to this system was "wifi", which seemed a little strange as I don't know anything about this topic. Even more odd was the fact that I had only four bookmarks tagged with "wifi", but apparently the number of bookmarks is not as important as what it is that you are actually bookmarking.

My current areas of expertise, as you can also see on the screenshot above: xml, mining, validation, geo, xslt, politics, wireless, wifi, vpn, blogs. Note that CollaborativeRank, for reasons unknown to me, split my "datamining" tag into the meaningless words "data" and "mining".

Note also that I have only one (1!) URL each under the tags "xslt", "wireless" and "vpn", yet I'm already deemed an expert on these topics. Isn't that strange? Not so, as you'll see.

When you think a little about it their algorithm seems quite straightforward. Extract from their FAQ:

The overall ranking [...] does provide a reward for people who provide helpful/timely URLs [...]

As an example, suppose Alice is in the 99th percentile with respect to the tag "crypto" while Bob is in the 70th percentile with respect to that tag. In this case, Alice would have greater influence on the search ranking for "crypto" than Bob.

The overall ranking [of a user] is computed in terms of these tag-specific user percentiles: the score for a user is the sum of his/her percentiles with respect to all the tags associated with that user.

So it's clear that this isn't simply an estimation of a user's area of expertise, but of his area of expertise relative to all other users. Let's paraphrase their cryptic lingo.

How to Get a High CollaborativeRank Listing

It's clear that this system is easy to abuse when you want to improve your ranking.

Tips on improving your CollaborativeRank listing:

Note that I did not actually try to test these hyoptheses though, so I could be wrong; a meaningful database is more important to me than an artificially high ranking.

Update: a couple of hours later I added links to my tags, and some additional explanations.

Next article:

Previous article:

Recent articles:


Comments are closed. You can contact me instead.