CNet is reporting that Microsoft and researchers have developed an algorithm that ranks search results not simply on link patterns, but on the browsing patterns of users. They call it BrowseRank. This is in contrast to Google’s PageRank algorithm that in part ranks search results on the links pertaining to content. You can read the original paper here: http://research.microsoft.com/users/tyliu/files/fp032-Liu.pdf
I think the researchers are wrong about their view that BrowseRank trumps PageRank.
Here are the key problems with this approach:
* BrowseRank is transparent. No one else but the people implementing the algorithm can see what the browsing patterns are. With links I can see within Google’s search results a bit of what’s going on. Links are public. Browsing patterns on the other hand aren’t. With links others can build and verify. How do we verify browsing patterns? All of the data would have to be public. No way does that make sense. I wouldn’t go for it. Links yes. Browsing patterns no.
* There’s a huge difference between taking content I select and showcase/post and collecting and organizing all my utterances (more akin to browse linking). The latter may be interesting–even to myself, but it’s not where we should be at right now in search.
* Imagine this algorithm implemented in China versus let’s say the US.
* If there’s one thing the web has taught us about public anonymity is that it is not to be trusted and BrowseRank is all about doing just this. What do I mean? Look at comments. Who do you trust more? A comment left and signed by Robert Scoble? Or one by “Anonymous?” This is exactly why services like Twitter and FriendFeed are finding more interesting conversations–because people attach their digital identities and reputations to them. If I put a link on my blog it’ll mean something because I put that link there. Other people can see it. Other people can confirm it (by linking) or not. If someone browses to a site, what does it mean? Is it a spam bot? An attack? A sign of agreement? Scoffing? Boredom? You’re going to have a trust ranking of the browsers. And where’s the archive of this behavior so it can be audited? And then who’s going to stand behind this? Will all Microsoft employees and partners make up this pool? Will all Slashdotters? Will it be opt in? See the problem?
* Browsing pattern analysis is like the Nielsen Ratings for websites. Leave it at that.
* With BrowseRank there is going to be spamming. It’s so obvious that spammers will develop bots and fake identities to taint the results. Just because spammers are going to design for it doesn’t mean it’s a bad idea, but here again, I’m concerned about the transparency of the implementation. The paper provides some details on why it thinks its algorithm can be used to prune spam, but obviously there’s a huge difference between applying an algorithm in vitro with one in the wild.
Now I have no problem with the algorithm being applied for marketing analysis or to Digg-like commenting sites and ranking their comments (comments written by people who really consumed the link content rise higher than those you just comment), or given a collection of XRank/Wikipedia content authors, just follow their browsing patterns and re-rank XRank. These are fine ways to use browse ranking in the wild. And then within contained environments (let’s say a company) it’s fine to rank content as I’ve blogged before by editing or consumption patterns. That makes sense. However, as a basis for all search results? Uhm.
I’m sorry, I don’t see how BrowseRank will scale well with human nature. In this alone, PageRank still trumps it.
[…] BrowseRank is worse than PageRank […]